In simple terms, we want to understand when, during the trading session, the S&P 500 is more likely to generate trend-following profits, and whether this pattern has changed over time.

This study was motivated by questions from our readers, who noticed, at least qualitatively, that SPY seemed to trend less consistently during certain parts of the trading session.

The goal of this article is to test this hypothesis empirically using our database and an ensemble of six different intraday trend-following models.

We begin by introducing the models. We then construct an ensemble portfolio to reduce model-specific noise in the results. Finally, we evaluate intraday seasonality over the full sample and investigate whether it has changed in recent years.

Intraday Trend-Following

Over the past few years, we have published several empirical studies on intraday trend-following strategies in U.S. equity markets.

Many of these ideas were first introduced by traders such as Toby Crabel, Linda Raschke, Larry Connors, Monroe Trout, and others in the early ‘90s. The basic intuition is simple: when price moves significantly away from a reference price level during the intraday session, the move may reflect a genuine imbalance between supply and demand.

Given our belief that capital can only move gradually through the market, we think these imbalances can persist throughout the trading session, allowing intraday trend traders to capitalize on them.

Our main contribution compared with previous studies is the accuracy and transparency of the empirical analysis. Previous studies relied on daily data and did not include explicit position management rules. In our papers, we used intraday data, incorporated simple position management rules, adopted realistic trading frequencies, and, in many cases, also shared the Python code we used to run the backtests.

For readers interested in the background, three useful references are our previous articles: Beat the Market, Can Day Trading Really Be Profitable?, and VWAP: The Holy Grail for Day Trading Systems.

In this article, we use an ensemble of intraday trend-following models that are very similar to the type of models we use in our own portfolios.

The reason we use an ensemble of models is simple: if we study only one model, the result may be driven by that model’s own rules. On the other hand, if we combine several models that all try to capture intraday trend continuation in slightly different ways, we have a better chance of isolating the underlying market behavior.

An Ensemble of Models

We use six intraday trend models. All of them trade SPY, the most liquid ETF tracking the S&P 500.

Below is a description of each model used in the study.

Read more

Trying to reverse-engineer the rules behind his program turned out to be a fascinating exercise, and we thank our readers for their support: without it, we wouldn't be able to keep this research publicly accessible.

Readers interested in the full details behind our replication pipeline, assumptions, and results can visit the original piece here.

Following several reader requests, we have extended the results of our best-fit replica through July 24 2026, offering an estimate of how Mulvaney might be positioned as the second half of the year begins.

As part of this update, we also review some of the year’s most significant trades, assess their contribution to portfolio P&L, and the lessons we can draw from them.

Importantly, this is not a disclosure of the fund's actual holdings: simply put, we take the specification we originally identified to best replicate Mulvaney's historical returns and present a rundown of its current exposures, without refitting or recalibrating the model on more recent data points.

At the end of this article, we also share a (free) interactive tool that lets you build and fine-tune your own version of the replica: read along to find out how you could win a one-year subscription to Concretum Research!

Inside Our Replica Model

As readers may recall, our original study was built around a basket of 39 futures markets spanning multiple asset classes and geographies: this remains our approximation of Mulvaney's investable universe, which has reportedly stayed unchanged since the fund launched in 1999.

The table below provides a snapshot of our best-fit replica’s holdings as of July 24, 2026. Since our model allows positions to be built through multiple tranches, each market is reported as long, short, or flat, with darker colors denoting a larger number of active tranches.

We additionally provide two measures of exposure: margin to equity and notional to equity. The former is a common metric used by CTAs to quantify how much capital is committed to supporting their positions, while the latter approximates the underlying economic exposure those positions represent.

Our replica is currently positioned in 19 of its 39 tradable contracts, with 8 long positions, 11 short positions, and the remaining 20 markets flat, making for an overall mild exposure level.

The clearest directional clusters appear in stock indices, interest rates, and currencies, with more fragmented positions across the broader commodities complex. Within those, the energy sector stands out, with long positions in Natural Gas (NG) and Low Sulphur Gasoil (GAS), alongside short exposure to metals, including both Gold (GC) and Silver (SI).

The same picture emerges when looking at net exposures by sector, where it is clear that grains, livestock, and softs currently hold only modest allocations.

We then isolate the replica’s best and worst trades initiated over the past year. Importantly, each tranche is treated as a separate trade, with its statistics evaluated independently.

The largest P&L contributor was the first (long) tranche in Brent Crude Oil (BRN), entered on January 29, 2026 and closed on June 19, 2026: against a backdrop of geopolitical tension, the trade captured much of the run-up in oil and adjacent energy markets.

Notably, this remained a remarkable trade even though the it gave back a significant portion of its open profit before exiting, as is typical of canonical trend-following programs. Such models are not meant to pick tops and bottoms; rather, they aim to capture the “belly” of a move, inevitably giving back part of their open P&L before a reversal eventually confirms the exit.

The worst P&L detractor was the second (short) tranche in E-mini S&P 500 futures. Entered on March 30, 2026, it was caught by an almost immediate rebound and exited on April 10, representing a fairly common trend-whipsaw: the signal triggered as the market broke lower, only for prices to quickly reverse and force liquidation of the position.

Interestingly, the contrast in duration between these two trades highlights one of trend following's most well-known characteristics: losers tend to be cut quickly, while winners generally need more time to develop and compound into meaningful trades.

Finally, we turn to market-level P&L to identify the instruments that contributed most and least to our replica's performance in 2026. Specifically, each market’s contribution is measured as its year-to-date P&L, expressed as a percentage of our replica AUM on the last trading day of 2025.

Consistent with our earlier observations, index futures were a modest source of drag this year, as short positions were caught out by equities pushing to all-time highs, with the notable exception of Asian equity markets (see MSCI Taiwan and Nikkei). The energy sector, by contrast, has been a solid contributor to P&L in 2026 so far.

Even though a degree of asymmetry between winners and losers is already visible here, it can become far more pronounced in other years. Take, for example, Cocoa (CC) in 2024, where its contribution within our replica portfolio was more than four times that of Coffee (KC), the second-largest.

This behavior, given how Mulvaney (presumably) structures his exposures and manages ongoing positions, is somewhat expected: often referred to as “old-school” trend, it represents one of the many areas where CTA managers often hold strongly diverging views.

For additional insights into position-sizing policies applied to trend programs, we suggest reading the following article.

What Does Today’s Positioning Tell Us?

Given our replica's overall mild exposure, one might be tempted to use this information to forecast future portfolio P&L, perhaps under the assumption that lean-allocation periods tend to precede “fresh”, rewarding trends.

This would only make sense if the number of active positions, margin usage or similar proxies reliably predicted subsequent trend persistance, a premise that, to our knowledge, has never been proven.

As such, we advise reading current exposures as a useful benchmark against other models already in use, rather than viewing them as allocation or trend-timing signals.

Build Your Own Mulvaney Replica

As a thank-you to our readers for their continued support, we’ve put together an interactive online tool that allows readers to build and backtest their own synthetic replica of Mulvaney’s fund, along with a chance to win a one-year paid subscription to the Concretum Substack.

Our original calibration relied on ordinary least squares (OLS), a widely adopted statistical method to estimate the best-fitting relationship between two return streams, but by no means the only valid one.

This time, we encourage readers to explore an alternative goodness-of-fit approach, one based on a purely qualitative and visual assessment, which may better capture the exceptional returns of Mulvaney’s fund over the past six years.

In practice, this means comparing replicas against Mulvaney’s actual track record and favoring the specification that best captures its shape and “character,” even if it isn’t the statistically optimal one.

As always, thank you for reading, we look forward to seeing what you build, and to hearing your thoughts in the comments.

If you found this article useful, we’d love to hear from you: let us know if you run trend-following strategies, and whether your current positioning looks consistent with our Mulvaney replica or takes a different view.

Research Transparency Statement
Every research idea, hypothesis, methodology, dataset, code, backtest, analysis, and conclusion presented in this article was independently developed by the Concretum Group research team. Artificial intelligence tools were used as productivity assistants to support the preparation of the final manuscript, including language refinement, grammatical review, and, where appropriate, the creation of tables, figures, and other visual elements. All research decisions, analyses, interpretations, and conclusions remain entirely those of the authors.

Disclaimer
This publication is provided by Concretum Group for informational, educational, and research purposes only. It does not constitute investment, financial, legal, or tax advice, nor a recommendation to buy or sell any security, instrument, strategy, or investment product. All investments involve risk, including possible loss of principal. Past performance, backtested performance, and historical analysis are not reliable indicators of future results. Readers should conduct their own research and consult qualified professionals before making investment decisions.

Full disclaimer: https://concretumgroup.com/disclaimer/

At some point, though, research usually asks for more than prices.

Finding a historical macro calendar is rarely the hard part. Finding one that is clean, consistent, easy to customize, and ready to merge into a backtest is. A lot of sources are built for looking at a webpage, not for dropping a file into a research pipeline, and turning them into something usable is quite time-consuming.

So, we opened up a small free tool we originally built for our own internal work. Select the events you care about (CPI, PPI, NFP, FOMC, GDP, Retail Sales, ISM, and others), pick a date range, and download a CSV you can plug straight into your workflow, no APIs needed. Same process we use whenever we want to study a strategy around scheduled macro events.

Macro Event Calendar

Subscribe now

Below, we’ll walk through what’s in the file, where to be careful with release times, and a simple use case on S&P 500 returns around macro days.

What’s Included in the Historical Economic Calendar

Pick any combination of:

FOMC, CPI, Non-Farm Payrolls, Unemployment, PCE, GDP, University of Michigan sentiment, Industrial Production, Initial Jobless Claims, Retail Sales, PPI, ISM Manufacturing, ISM Services, and Conference Board Consumer Confidence.

Then export in one of two layouts:

• Long: one row per event occurrence. Easy to filter and join.

• Wide: one row per trading day, with a 0/1 column per event. This is usually what you merge into a backtest.

The long file layout includes stable identifiers and a few research-friendly flags:

One small but important detail: the calendar includes future scheduled dates, not just past ones. That’s the difference between an archive and a tool you can actually plan around.

Assumed times: useful, but handle with care

Release dates are the core of the file. Intraday times are harder, and we do not want to pretend otherwise.

The tool can optionally fill assumed_time_et from the conventional U.S. publication schedule. That option is off by default. The column is named assumed_time_et on purpose, not time_et. Treat those stamps as reconstructed estimates, and verify them before you rely on them for anything intraday.

Two reasons for the caution:

1. The convention is not stable. The FOMC statement moved from 14:15 to 14:00 ET in 2013. Shutdowns and schedule changes have also shifted prints historically.

2. One timestamp is often incomplete. An FOMC day can include the statement, a press conference later, and projections at some meetings. Michigan has a preliminary and a final. The assumed time is for the primary release only.

If you only need dates for a daily or overnight study, leave assumed times off. The date spine alone is already useful.

Using the Historical Economic Calendar for Event Studies

One simple use case for this calendar is to ask how U.S. equity returns behaved on days associated with different scheduled macro releases.

We join the wide CSV to the S&P 500 total return index and compute the average daily return, in basis points, conditional on each event being scheduled. The sample runs from 1990 to today, and the dashed line in the first chart represents the unconditional full-period average, about 4.8 bps.

FOMC days stand out with the highest average return, followed by Consumer Confidence, ISM Manufacturing, and ISM Services. GDP and Industrial Production also sit above the unconditional benchmark. Events such as PCE, PPI, and jobless claims are closer to the full-sample average.

A second cut moves from individual events to event density: how many macro releases from our list fall on the same trading day.

Days with no scheduled macro events delivered only a few basis points on average, whereas days with at least one or two events showed higher average returns. The increase becomes more visible from three events upward, and the top bucket (five or more) printed above 30 bps, though that group has only 15 observations.

This is indicative of higher average returns on denser information days: it is a starting point for research, but not a causal conclusion and not a trading rule.

Beyond daily index returns, the same file can be used to study how intraday strategies behave around key releases. For example, whether an intraday trend signal behaves differently on FOMC, CPI, NFP, ISM, or GDP days, when information shocks and order-flow imbalances are more likely.

That connects to our earlier article on an intraday trend strategy, where the rules are fully disclosed and results are broken out around selected macro dates.

Future Improvements

The CSV is a calendar, not a “surprise” database. It tells you when the release was scheduled. It does not attach the first print, the revised value, or the consensus miss.

That is the natural next layer: values on those dates, ideally first print next to the revised figure, and eventually surprise versus consensus. The date spine comes first.

If you already keep a hand-built event list for backtests, this is meant to replace that chore. Select the releases, export long or wide, merge it into your returns, and run the cut you actually care about.

Macro Event Calendar

Subscribe now

Verify assumed times before you use them intraday, and if you find something cleaner than average return on event days, send it our way.

Research Transparency Statement
Every research idea, hypothesis, methodology, dataset, code, backtest, analysis, and conclusion presented in this article was independently developed by the Concretum Group research team. Artificial intelligence tools were used as productivity assistants to support the preparation of the final manuscript, including language refinement, grammatical review, and, where appropriate, the creation of tables, figures, and other visual elements. All research decisions, analyses, interpretations, and conclusions remain entirely those of the authors.

Disclaimer
This publication is provided by Concretum Group for informational, educational, and research purposes only. It does not constitute investment, financial, legal, or tax advice, nor a recommendation to buy or sell any security, instrument, strategy, or investment product. All investments involve risk, including possible loss of principal. Past performance, backtested performance, and historical analysis are not reliable indicators of future results. Readers should conduct their own research and consult qualified professionals before making investment decisions.

Full disclaimer: https://concretumgroup.com/disclaimer/

This principle lies at the heart of the Capital Asset Pricing Model (CAPM), one of the most influential models in financial economics. According to the theory, stocks with higher systematic risk should compensate investors with higher long-term returns.

Yet among the many anomalies documented in empirical finance, few challenge this principle more directly than the low-volatility anomaly.

Over the past several decades, researchers have repeatedly found that stocks with the highest historical volatility have not only delivered lower risk-adjusted returns, but often lower absolute returns as well. In other words, investors who accepted the greatest amount of risk were frequently the least rewarded.

In our previous articles, we introduced the concept of decile portfolios and showed how they can be used to study well-known equity characteristics such as Value. In this article, we apply the same methodology to another of the most robust anomalies in equity markets: Low Volatility.

The results are surprisingly difficult to reconcile with traditional asset-pricing theory.

The Low-Volatility Puzzle

Back in 1972, Fischer Black observed that the empirical Security Market Line (SML) was too flat. High-beta stocks earned lower average returns than predicted, while low-beta stocks performed better than the CAPM suggested.

Decades later, Andrew Ang and his co-authors documented a closely related result: stocks with high idiosyncratic volatility tended to earn abnormally low future returns (link). Subsequent research showed that this effect was not limited to the United States, but appeared across several developed equity markets.

The chart below applies the same decile methodology used in our previous articles. At the end of each month, stocks are ranked by their historical variance and divided into ten portfolios. Decile 1 contains the lowest-variance stocks, while Decile 10 contains the highest-variance stocks.

The results are quite surprising.

There is no positive relationship between risk and return. In fact, the stocks with the highest variance perform worse than the rest of the market.

• The lowest-variance portfolios generate positive CAPM alphas.

• As variance increases, performance gradually deteriorates.

• The highest-variance decile ends up producing a strongly negative alpha.

This finding clearly contradicts the basic idea behind traditional asset pricing theory, which suggests that investors should earn higher expected returns for taking more risk. In our sample, we observe exactly the opposite.

The same conclusion emerges when looking at Sharpe ratios. The highest-volatility decile produces by far the worst risk-adjusted performance over the whole sample period. To put some numbers on it, the lowest-variance decile achieved a Sharpe ratio of 0.71, while the highest-variance decile reached only 0.03.

These results help explain why the low-volatility anomaly is often considered one of the clearest contradictions of modern financial theory.

Why Does the Anomaly Exist?

There is no single universally accepted explanation. The most common interpretations combine institutional constraints, behavioral biases, and limits to arbitrage. Let’s explore some of them…

Leverage constraints. Many institutional investors cannot use leverage, so they overweight high-beta stocks instead of leveraging low-beta stocks.

Behavioral biases. Investors are drawn to high-volatility, “lottery-like” stocks, often overpaying for them.

Short-selling constraints. The worst-performing high-volatility stocks are often difficult to short, allowing mispricing to persist.

How to Trade the Low Volatility Anomaly

The low-volatility effect can be transformed into a long-short systematic strategy by buying stocks with low volatility while shorting stocks with high volatility.

A “volatility-neutral” implementation could be constructed as follows:

• Go long the lowest-volatility decile

• Short the highest-volatility decile

• Scale the two legs to the same target volatility (20% p.a.).

Over the full sample (1963–2026), this simple strategy would have delivered a 26% compound annual growth rate (CAGR), with a Sharpe ratio of 0.87, an annualized CAPM alpha of approximately 21%, and virtually zero market beta.

However, the strategy is far from risk-free. Low-volatility portfolios have historically experienced prolonged periods of underperformance, particularly during sharp market rebounds when speculative, high-beta stocks can move aggressively.

The equity curve below highlights several notable drawdown episodes, including 1966, the late-1990s technology bubble, and the post-COVID rebound. Some of these drawdowns exceeded 50%, illustrating that even one of the most persistent anomalies in equity markets can experience long and painful periods of adverse performance.

The Volatility Anomaly in 2026

The charts shown throughout this article are based on equally weighted decile portfolios, which provide the cleanest representation of the academic low-volatility anomaly.

In practice, however, many investors implement factor portfolios using value-weighted constituents.

Interestingly, 2026 has so far provided a useful reminder that factor premiums can experience meaningful short-term deviations.

Using a value-weighted construction, the highest-volatility decile (D10) has returned an impressive +83% YTD (through the end of May), while the lowest-volatility decile (D1) has declined −5.3%. Consequently, a traditional value-weighted low-volatility long-short portfolio has experienced a significant drawdown this year.

This divergence is largely a consequence of the exceptional market concentration observed in 2026. Value-weighted portfolios naturally concentrate capital in the largest companies, leaving them heavily exposed to the extraordinary performance of a handful of high-volatility mega-cap technology and semiconductor stocks that have dominated this year's AI-driven rally. Meanwhile, many of the largest low-volatility mega-cap companies have lagged the broader market. The resulting performance gap between the two deciles is therefore substantially larger than under an equally weighted construction.

In fact, the picture is noticeably different using the equally weighted portfolios shown throughout this article. In this case, the highest-volatility decile is up +17.2% YTD, while the lowest-volatility decile has gained +5.4%. An equally weighted long-short implementation with volatility targeting approach is therefore down only about 1% YTD, despite one of the strongest rallies in the most volatile growth stocks in recent years.

This highlights an important lesson: portfolio construction matters. Choices such as equal versus value weighting, volatility scaling, and position concentration can materially affect short-term performance, even when the underlying long-term anomaly remains intact.

It also raises a broader question:

Should investors maintain unconditional exposure to a factor even when its performance trend has deteriorated?

In a recent article, we explored this idea by applying cross-sectional and time-series momentum techniques across equity factors. A momentum-based allocation would likely have reduced or exited exposure to the low-volatility factor relatively early during its 2026 deterioration, potentially limiting the impact of the drawdown.

For readers interested in this approach, see the following article where we examine how momentum can be used to allocate dynamically across equity-factor premia.

More detailed and up-to-date comparisons between equal-weighted and value-weighted factor portfolios are available through Concretum Factor Tracker.

Conclusion

The low-volatility anomaly is probably one of the most interesting contradictions in finance. According to the CAPM, investors should earn higher expected returns by taking more risk. However, decades of empirical evidence suggest exactly the opposite: the more risk you take, the lower your long-term returns tend to be.

Of course, this does not mean that investing in low-volatility stocks is a free lunch. Like any other equity factor, it can go through long periods of weak performance, sometimes for several years. As we have seen during 2026, portfolio construction, market concentration, and factor timing can have a much larger impact on short-term results than many investors expect, even if the long-term premium is still there.

If you would like to explore the anomaly in greater depth, we strongly suggest to read Pim van Vliet’s excellent book High Returns from Low Risk.

All the charts used in this article were generated using Factor Tracker, a FREE tool developed by the our team in Concretum.

If you found this article useful, feel free to leave a comment or contact us by direct message or at info@concretumgroup.com.

Disclaimer
This publication is provided by Concretum Group for informational, educational, and research purposes only. It does not constitute investment, financial, legal, or tax advice, nor a recommendation to buy or sell any security, instrument, strategy, or investment product. All investments involve risk, including possible loss of principal. Past performance, backtested performance, and historical analysis are not reliable indicators of future results. Readers should conduct their own research and consult qualified professionals before making investment decisions.

Full disclaimer: https://concretumgroup.com/disclaimer/

While our findings are consistent with those of Panjabi, Bordigoni, and Buchanan (2026), they should not be viewed as direct confirmation that such actively managed factors can be traded directly as constructed.

Implementing long-short equity portfolios presents several practical challenges, and although transaction costs remain an important consideration, today we specifically focus on short-selling, which constitutes a significant (and less discussed) pain-point.

In particular, we present an alternative implementation of our original study, built to offer a more practical (and scalable) version of the factors we previously examined.

Our results suggest that momentum effects persist under alternative factor-construction techniques that, over the last twenty years, would have still proven effective in isolating exposure to the best-performing equity styles.

As the numbers above highlight, our actively managed framework helps revive an otherwise stagnant blend of factor portfolios.

For investors seeking equity-style exposures that are easier to trade in practice, we believe this article offers some valuable insights.

Read more

The appeal of these strategies is not only their standalone return profile. In many cases, their real value comes from the fact that, by construction, they deliberately take orthogonal exposure from a passive long-only equity allocation. This can translate into a significant benefit in terms of correlation control and alpha generation.

Our public work on this topic started with Beat the Market: An Effective Intraday Momentum Strategy for S&P500 ETF (SPY), published in 2024, where we showed how intraday demand and supply imbalances can be transformed into a systematic trend-following framework.

Most of the models we have presented so far share a common design choice: they use recent realized volatility to define the intraday breakout levels and, in many cases, to size positions dynamically.

That choice is simple, transparent, and hard to dislike. But it is also backward-looking.

This naturally leads to a question we wanted to test more carefully:

Can implied volatility improve the precision with which we define breakout levels and size intraday trades?

In this article, we introduce the full rules behind an effective intraday volatility breakout strategy on SPY that compounds at 17% per year (net of cost) over the backtest window, while showing an annualized alpha of ~18% and a beta of -0.06 versus SPY.

We then replace historical volatility with VIX readings when calibrating the bands and assess whether option-implied volatility can improve the effectiveness of the strategy. We will show that VIX data can be used more effectively when we avoid introducing unnecessary lag in the volatility input.

Moreover, we ask whether implied volatility can be used for another part of the model: position sizing.

Finally, we assess whether VIX information becomes more useful during macro-announcement days such as FOMC, CPI, ISM, Initial Jobless Claims, and Nonfarm Payrolls.

There are several useful lessons in this article. Enjoy your read.

Read more

The authors show that cross-sectional momentum techniques can be successfully applied across equity-style factors, potentially providing meaningful diversification relative to traditional time-series approaches.

In practical terms, an investor who already holds a conventional trend-following allocation across asset classes may benefit from adding a more orthogonal source of return: going long the strongest-performing equity factors while shorting the weakest-performing ones.

The chart below illustrates the equity curve of the style-trend portfolio presented by Man AHL.

Needless to say, our team found the piece sufficiently insightful to warrant further investigation.

In this article, we extend the analysis using Professor Kenneth French’s publicly available data library for U.S. equity factors.

We divide the sample into two adjacent periods: 1985–2005 and 2006–2026. The latter closely matches the period studied by Man AHL, while the preceding 20 years provide a useful historical comparison. The earlier sample can also be interpreted as broadly in-sample-like, since many of the underlying anomalies were discovered or popularized using data from that period.

As is often the case with studies based on academic factor portfolios, the results should not be interpreted as a ready-to-trade investment strategy. In our view, however, they highlight (and confirm) several findings that may be valuable when constructing actively managed, factor-based equity portfolios.

Factor Portfolio Construction

For our analysis, all data are sourced from Professor Kenneth French’s Data Library, one of the most widely used standards in academic factor research. The portal hosts hundreds of datasets covering portfolios formed on firm characteristics, industry groupings, and other classifications.

Readers who are new may find the Concretum Factor Tracker useful. It is a free tool we built to conveniently collect, track and visualize several US equity factor portfolios from the French library.

Factors are typically constructed as long-short exposures to decile portfolios. For each characteristic (such as value, momentum, or profitability) stocks are sorted into ten portfolios, or “deciles,” according to their ranking on the selected signal.

One important design choice concerns the weighting scheme within each decile portfolio. Stocks can be equally weighted, assigning the same weight to every constituent, or value weighted, with larger companies receiving proportionally greater weights based on market capitalization. Since both approaches are widely used in academia and practice, we examine them separately throughout the study.

As a convention, we identify D10 as the decile that historically earns the documented premium and D1 as the opposite extreme, flipping the source labels where necessary: each factor is therefore constructed as a long position in D10 and a short position in D1.

A second design choice concerns how the two legs are combined into a long-short factor portfolio.

One common approach is dollar-neutral weighting, where each decile carries 100% notional exposure, giving the combined portfolio 200% gross exposure.

Another widely used approach is volatility scaling, under which each leg is resized periodically to target a predefined level of volatility. Because the realized volatility of D10 and D1 can differ substantially, scaling the two legs independently prevents the more volatile side from dominating the portfolio’s total risk.

Unless otherwise stated, this study adopts the second approach and assigns the same volatility target to both legs (see “Momentum Has Its Moments” by Barroso and Santa-Clara (2015) for additional discussion).

Our study relies on monthly data and covers fifteen anomalies made available within the French Data Library for US markets. Full details are provided in the table below.

For all factors, both the long and short legs are sized to target 20% annualized volatility, using a 12-month rolling standard deviation of log monthly returns as our volatility proxy. Formally, for each month t:

The monthly return of each factor portfolio is then computed as:

Cross-Sectional Momentum

The Panjabi-Bordigoni-Buchanan article provides a useful starting point for defining our cross-sectional signal:

While the exact formulation behind their relative-strength measure remains proprietary, we can start by simply defining each factor’s momentum (M_i,t) as its return over the prior 12 months:

A useful feature of constructing factors from volatility-scaled legs is that their trailing returns are directly comparable, without requiring further normalization. Cross-sectional sorts are most meaningful when applied to like-for-like portfolios: ranking raw returns across portfolios with very different risk levels could otherwise distort the ordering. Because all fifteen factors use the same volatility target for each leg, this issue is materially reduced.

Results

Our simulation starts in January 1985 and runs through May 2026, a span we split into two 20-year samples: 1985-2005 and 2006-2026.

We begin by testing whether the factor-momentum rank observed at each month-end (t) contains information about returns in the following month (t+1). For each rank, we calculate the average next-month returns separately by time sample and weighting scheme (equal- and value-weighted decile portfolios). The chart below reports the results.

The results appear encouraging: the average next-month returns decline steadily as ranks move from 1 (the strongest relative-strength factor) to 15 (the weakest). The relationship is not perfectly monotone (with a few middle-ranks step out of line) but the overall slope is rather clear, and it shows up for both the equal-weighted (EW) and the value-weighted (VW) constructions, in both 20-year samples.

To translate this rank-return relationship into a simple strategy, each month we buy the three strongest factor portfolios and sell short the three weakest, based on their prior 12-month returns. The nine middle-ranked factors receive no allocation.

The chart below compares this strategy against a “passive” benchmark that holds all fifteen (volatility-sized) factors in equal weights and rebalances monthly. Within the figure, rows separate the weighting scheme (EW above, VW below) and columns split the two 20-year samples.

Notably, before 2006, the active portfolios fail to outperform the unconditional benchmark. We believe this result is largely a consequence of evaluating the strategy over the original factor-discovery period. By construction, the factors documented in the academic literature were identified precisely because they delivered positive returns during this sample. As a result, even the relatively weaker factors assigned to the short side often continued to exhibit positive returns.

This creates a structural headwind for the long-short implementation: the short leg is systematically positioned against factors that still possess positive in-sample drift. By contrast, the benchmark remains long every factor and therefore captures the full benefit of this broad-based appreciation.

After 2006, the picture changes materially. This period is arguably the more informative sample because many canonical factors had entered their genuine out-of-sample, post-publication phase. Published anomalies often experience substantial performance decay thereafter, whether because of arbitrage, changing market conditions, or the possibility that some original findings reflected data-mining artifacts.

In such an environment, an unconditional allocation across all factors becomes less compelling. The cross-sectional momentum strategy instead reallocates capital toward factors that continue to display relative strength and takes short exposure to those whose premia have faded. As a result, the active portfolios continue compounding, while the benchmark advances much more slowly or, in the value-weighted case, scarcely at all.

The same conclusion emerges from risk-adjusted returns. Sharpe ratios are recomputed within each 20-year sample for both the equal-weighted and value-weighted portfolios and are shown below.

Over 1985–2005, the equal-weighted benchmark produced a remarkable Sharpe ratio of 2.33, compared with 1.18 for the active portfolio. Under value weighting, the corresponding figures were 1.09 and 0.78. Over 2006–2026, however, the pattern reverses: the benchmark falls to 0.63 under equal weighting and just 0.10 under value weighting, while the active portfolio remains at 1.18 in the equal-weighted case and improves to 0.82 in the value-weighted case.

How About Time-Series Momentum?

Having found robust evidence of cross-sectional momentum effects, we additionally check for time-series (i.e. trend) effects. Here the timing signal (s) no longer depends on the monthly cross-sectional sort: each factor (i) is simply traded long or short according to the sign of the 12-month momentum feature calculated earlier. Formally:

The simulated results are presented below, using the same benchmark described earlier for comparison.

The time-series signals tell a similar story. The active factor portfolio trails the always-long D10-minus-D1 basket during the first sample but navigates the post-2006 environment much more effectively. As before, the result holds under both equal-weighted and value-weighted factor constructions.

Practical Implications

The portfolios presented throughout this article should primarily be viewed as academic building blocks rather than directly tradable investment strategies. Nevertheless, we believe the underlying findings are highly relevant for systematic investors seeking practical ways to dynamically allocate capital across equity factor premia.

In practice, however, implementing these strategies is far from straightforward. Academic factor portfolios require trading and, in particular, shorting hundreds of individual stocks. In reality, many names may not be available to borrow, and managing them introduces significant scalability and execution challenges.

Our follow-up article explores a framework for translating these rotational factor-momentum models into more investable portfolios: a natural next read if you're interested in the practical implementation of these ideas. Click the banner below to subscribe and receive it upon publication.

Conclusion

Overall, our results confirm the findings of Panjabi, Bordigoni, and Buchanan (2026) with a meaningful degree of robustness.

Across our factor universe, both cross-sectional and time-series momentum effects appear in each of the two 20-year samples and under both equal-weighted and value-weighted decile constructions.

However, the most interesting finding is not that factor momentum exists, but when active factor selection becomes most valuable.

During the original discovery period, nearly all documented factors exhibited positive drift by construction. In that environment, a long-short allocation faces an inherent disadvantage because its short leg is positioned against factors that were (obviously) producing positive returns. An unconditional, always-long allocation can therefore be difficult to beat.

Once these factors enter their post-publication period, the economics change. Some premia weaken, some become crowded or arbitraged away, and others may reveal themselves to have been partly sample-specific. In that setting, unconditional exposure becomes substantially less attractive, while momentum-based allocation offers a systematic way to bet on factors that continue to work and against those whose premia have weakened or reversed.

Viewed against the broader “factor zoo,” this distinction is important. The term, popularized by John Cochrane, describes the proliferation of empirical factors beyond the Capital Asset Pricing Model (CAPM). By some estimates, several hundred factors have now been published, raising persistent concerns about data mining and weak out-of-sample performance (Harvey, Liu, and Zhu, 2016).

In such a landscape, momentum-based approaches (whether cross-sectional or time-series) for factor investing deserve further study not merely as standalone anomalies, but as practical allocation tools for navigating a large and evolving universe of equity styles.

If you found this article useful, feel free to leave a comment or contact us by direct message or at info@concretumgroup.com.

Disclaimer
This publication is provided by Concretum Group for informational, educational, and research purposes only. It does not constitute investment, financial, legal, or tax advice, nor a recommendation to buy or sell any security, instrument, strategy, or investment product. All investments involve risk, including possible loss of principal. Past performance, backtested performance, and historical analysis are not reliable indicators of future results. Readers should conduct their own research and consult qualified professionals before making investment decisions.

Full disclaimer: https://concretumgroup.com/disclaimer/

Whether driven by central-bank support, structural inflows, or investor psychology, market participants have become increasingly conditioned to treat short-term weakness as an opportunity rather than a threat.

Yet one surprisingly practical question receives far less attention:

How should the dip actually be bought?

Should traders buy immediately after the signal appears? Wait until the next session? Or demand a deeper selloff before committing capital?

In this article, we revisit one of the best-known mean-reversion concepts in quantitative trading and compare several execution policies across more than 70,000 pullback trades in S&P 500 stocks between January 2010 and May 2026.

While the underlying signal remains identical throughout the study, we find that average profitability varies dramatically depending solely on how positions are entered, ranging from roughly 30 basis points to nearly 180 basis points per trade.

More importantly, the results suggest that there is no universally optimal entry policy. The preferred execution method ultimately depends on the role the strategy is expected to play within a broader portfolio.

For an investor seeking a small number of high-conviction opportunities, a highly selective entry policy may be preferable. For a manager building a dedicated mean-reversion portfolio, however, the ability to generate enough trades to keep capital consistently deployed may matter more than maximizing expectancy on every individual position.

Read more

The positive response from readers naturally led us to a broader question:

Can the same idea be extended into a diversified portfolio of sector ETFs?

In this article, we apply the exact same framework to the complete universe of eleven SPDR Sector ETFs, creating an actively managed portfolio that dynamically allocates capital only when statistically significant reversal opportunities emerge.

The results are encouraging.

Over the period from June 2018 through early July 2026, the strategy compounds at 18.5% per year after transaction costs, comfortably outperforming a passive equal-weight benchmark, which returns 12.6% annually over the same period.

At the same time, it improves the portfolio’s Sharpe ratio from 0.75 to 1.33, reduces the maximum drawdown from -36.3% to -14.7%, and achieves these results while deploying only 58% of its capital on average, leaving a substantial portion unallocated throughout much of the sample.

For simplicity, we do not account for the interest that this idle cash could have earned, meaning the reported returns are likely conservative.

In the remainder of this article, we unpack the full methodology behind the strategy, examine the empirical behavior supporting the signal, and show how the individual trades are combined into a diversified portfolio. We then evaluate its performance after transaction costs, capital usage, and resilience across different market environments.

Read more

Value investing has been one of the most enduring investment philosophies, pioneered by Benjamin Graham and later popularized by Warren Buffett.

At its core, value investing is based on the idea that markets are not always efficient, and mispricings occur, allowing investors to buy stocks at a discount to their intrinsic value.

Graham, in Security Analysis (1934) and The Intelligent Investor (1949), emphasized the importance of buying stocks with a margin of safety, meaning stocks that trade below their fundamental value as measured by earnings, book value, or cash flows.

Buffett, Graham’s most famous student, refined this approach by incorporating qualitative factors such as economic moats and strong management teams.

The Academic Perspective: Quantifying Value Investing

Academia has long sought to quantify the stock selection process used by successful value investors. A key breakthrough came with Eugene Fama and Kenneth French (1992, 1993), who introduced the High Minus Low (HML) factor in their three-factor model. Their research showed that stocks with high Book-to-Market (B/M) ratios (a proxy for cheap valuations) tended to outperform stocks with low B/M ratios over long periods.

Subsequent studies expanded on this by testing different valuation metrics such as price-to-earnings (P/E), price-to-cash flow (P/CF), and enterprise value-to-EBITDA (EV/EBITDA).

Across different markets and time periods, the evidence consistently supported the existence of the value premium, whereby cheap stocks tended to generate higher risk-adjusted returns than expensive stocks.

Empirical Evidence: Performance of Value Portfolios

To visualize the strength of the value premium, the following chart presents the alpha of decile portfolios relative to the CAPM model, sorted by Book-to-Market ratio, from 1926 to 2026. The decile with “cheap” value stocks is D10.

The chart reveals a nearly linear relationship between monthly alphas and value deciles, demonstrating that stocks in the highest B/M decile (Decile 10, the most undervalued) generated the highest excess returns, while those in the lowest B/M decile (Decile 1, the most expensive) performed the worst.

This pattern is also evident in the equity curve below, where Decile 10 (high-value stocks) consistently outperforms Decile 1 (low-value stocks), reinforcing the historical advantage of investing in undervalued companies.

Why Does the Value Premium Exist? Risk vs. Behavioral Explanations

The persistence of the value premium has sparked debate on whether it is a compensation for risk or a market inefficiency due to behavioral biases.

Risk-Based Explanation

One view is that value stocks are fundamentally riskier investments, and their higher returns compensate for this additional risk. Value stocks often belong to companies that are financially distressed, operate in declining industries, or have uncertain growth prospects. During economic downturns, these stocks tend to suffer more than their growth counterparts. Fama and French (1993) originally supported this explanation, arguing that value stocks load on risk factors not captured by traditional CAPM.

Another argument for the risk-based explanation is that value stocks are more sensitive to macroeconomic conditions, such as changes in interest rates, inflation, and credit cycles. Investors may demand a premium for holding stocks that are more exposed to these risks.

Behavioral Explanation

However, alternative behavioral theories emerged, particularly in Lakonishok, Shleifer, and Vishny (1994), which argued that the value premium arises from investor overreaction and mispricing rather than risk. Investors tend to suffer from representativeness bias, a situation where they naively extrapolate past earnings growth rates too far into the future.

The chart below, taken from the great book Quantitative Value by Wesley Gray, Ph.D. and Jack Vogel, shows the average past five-year earnings growth rates for each decile portfolio formed on the Book-to-Market metric.

It is evident that the more a stock is classified as “Value” (i.e., higher book-to-market ratio), the lower its recent earnings growth. The relationship is almost perfectly linear.

Now the question is:

Do value stocks with low earnings growth rates continue to exhibit low earnings growth in the future?

The answer is provided in the figure below and is a decisive NO!

Value stocks in the top decile tend to have the highest future earnings growth rates, making investors not only buy at a low price relative to book value but also buy stocks that, on average, tend to have higher earnings growth in the future. In some sense, this chart suggests that value investing exposes investors to stocks that will become some of the fastest-growing stocks in the future.

Challenges of Value Investing and the Role of Momentum

One challenge of value investing is that it can experience prolonged periods of under-performance relative to the market. Investors with short-term investment horizons may find it difficult to stick with value strategies, as these drawdowns can last for years. A notable example occurred in the late ‘90s when value stocks dramatically underperformed the broader market. During this period, often referred to as the dot-com bubble, even some of the most successful value investors suffered significantly. For instance, Julian Robertson’s Tiger Funds saw its AUM decline sharply, while Barron’s magazine questioned whether Warren Buffett had lost his magic touch.

This has led to the search for ways to enhance value investing, and one of the most effective methods identified by both academics and practitioners is combining it with momentum investing.

Well-known academics and practitioners, such as Cliff Asness, Tobias Moskowitz and Lasse Pedersen, have shown that adding a momentum leg to a value strategy significantly improves risk-adjusted returns. The benefit of this combination is pervasive, not only in equities but also across other asset classes, including bonds, commodities, and foreign exchange.

In our own trading, for example, we also exploit the orthogonality between value and momentum signals when trading volatility. This diversification effect helps smooth out drawdowns and provides more consistent returns across different market environments.

Conclusion

Despite its historical success, value investing has faced challenges in recent years, particularly post-2008, when growth stocks dominated. Some argue that changing market structures and the rise of intangible assets have diminished the effectiveness of traditional value metrics. Nevertheless, value investing remains a fundamental strategy for investors seeking long-term outperformance, supported by both empirical data and the principles laid out by Graham and Buffett.

Interestingly, in 2020 and 2021, value stocks experienced a strong resurgence, with the top decile of value stocks returning 53% in 2020 and an impressive 91% in 2021 when constructed using an equally weighted methodology. This remarkable rebound highlights the cyclical nature of the value premium and underscores the importance of maintaining a long-term perspective when investing in value strategies.

While the debate over whether the value premium is a compensation for risk or an anomaly driven by behavioral biases continues, its long-term persistence suggests that patient investors can still benefit from applying value principles in a disciplined manner, potentially enhanced by the inclusion of momentum investing.

📊📈📉 The first 2 charts used in this article were generated using a FREE tool developed by the Concretum|Research team.

If you found this article useful, feel free to leave a comment and reach out via direct message or email at info@concretumgroup.com for any questions.

Disclaimer
This publication is provided by Concretum Group for informational, educational, and research purposes only. It does not constitute investment, financial, legal, or tax advice, nor a recommendation to buy or sell any security, instrument, strategy, or investment product. All investments involve risk, including possible loss of principal. Past performance, backtested performance, and historical analysis are not reliable indicators of future results. Readers should conduct their own research and consult qualified professionals before making investment decisions.

Full disclaimer: https://concretumgroup.com/disclaimer/

Our results indicate that, in its plain-vanilla form, the strategy fails to generate meaningful net-of-fee returns: commissions and market impact erode the gross edge almost entirely, and participation on virtually every trading session leaves the strategy fully exposed to cumulative trading costs.

Rather than abandoning the ORB concept altogether, however, we ask a different question:

Can a handful of simple, economically motivated improvements transform an otherwise mediocre strategy into an attractive systematic trading framework?

Specifically, we investigate whether a combination of intuitive trade-selection filters, improved trade management and dynamic position-sizing techniques can materially improve both returns and risk-adjusted performance without altering the core ORB philosophy or relying on extensive parameter optimization.

Our findings suggest that the answer is yes!

As illustrated in Figure 1, these enhancements increase the strategy's CAGR from 1.0% to 12.1%, improve its Sharpe ratio from 0.14 to 0.84, and raise annualized alpha from an economically modest and statistically insignificant 1.75% to a highly significant 12.80%, all while remaining robust under conservative assumptions for commissions and market impact.

The edge lies not in inventing a new ORB signal, but in learning how to trade the existing one more intelligently.

In the remainder of this article, we explain each enhancement individually, discuss the economic intuition behind it, and quantify its incremental contribution to the strategy's overall performance.

Read more

From investors debating whether the recent drop in software-related names makes for a compelling long-term opportunity…

…to more active market participants weighing the odds of a correction in semiconductors after an essentially unprecedented run…

…to swing traders shifting their attention to biotech, which has recently set a fresh multi-year high after a long stretch of consolidation.

As a team that manages proprietary capital through quantitative frameworks, market views and commentary sit outside of what we deal with day to day, but every now and then they can spark some interesting ideas.

Inspired by the dispersion in sector returns we have recently witnessed, today we present a trading strategy that seeks to exploit momentary price dislocations, in a way that remains attractive net of conservative transaction-cost assumptions.

As a case study, we apply it to three industry ETFs that capture some of the sector-rotation “debate” mentioned earlier: the iShares Semiconductor ETF (SOXX), the iShares Expanded Tech-Software ETF (IGV), and the iShares Biotechnology ETF (IBB).

We show how a portfolio constructed by actively timing these ETFs has generated a net-of-fee CAGR of 24.6% and a Sharpe ratio of 1.17, while keeping maximum drawdown at ~20%.

In addition to the rules behind the signal and our portfolio-construction approach, this piece gives us a chance to touch on a delicate topic that, in our opinion, is central to the analysis and implementation of many short-term trading strategies.

Read more

If you have ever worked on quant-trading desks, or been involved in advisory work, you already know that portfolio managers and institutional clients are always keen to know whether a signal can support a certain amount of capital in production and real-world trading settings.

With this new follow-up piece, we would like to present some additional insights on how fees and market impact can be used to estimate the capacity of a trading strategy. Using an intraday mean-reversion signal as a case study, we hope to offer a useful framework for how, as quants, we can try to tackle the “capacity dilemma”.

Dissecting Capacity

In our experience, capacity is best understood as a three-way problem. More specifically, we believe it can be broken down into three sub-questions:

The first question relates to an upper-bound problem, where we want to determine the maximum amount of capital a signal can sustain before being eaten away by costs. Given what we already discussed at length in our previous article, we expect market impact to be the predominant cost driver in this case.

The second question poses a classic optimization problem, where we seek the ideal amount of capital that minimizes the total impact of frictions (comprised of fees and slippage), allowing us to capture the highest possible (net-of-fee) risk-adjusted returns.

Last but not least, the third question constitutes a lower-bound problem. As unusual as it might sound, there is such a thing as trading a signal with too little capital: knowing that fees are often priced with hard minimum-commission thresholds, committing very limited capital to a given strategy means many trades could hit that commission floor, disproportionately eroding realized P&L.

Crucially, none of these questions matters more than the others: they are all relevant, as market participants can face widely different constraints and limitations.

A Valid Framework

Now that we have established that the capacity problem is more multi-faceted than it might appear at first glance, we can lay out a framework to answer all three key questions (one we have adopted many times in our practice) summarized in the following steps:

1. Define a set of reference AUMs, which act as discrete anchor points at which the signal must be evaluated, spanning the range of possibile capital allocations we wish to study,

2. Simulate how the signal responds (net-of-fee) across those reference AUMs, through a continuous cost-normalization procedure,

3. Keep track of how risk-adjust metrics (i.e. Sharpe ratios) vary as the reference AUM progressively increases.

While the first step is trivial and rather arbitrary, we think the second one deserves more attention, namely for its cost-normalization procedure.

From our previous article, we learned that transaction costs do not scale linearly with traded notional, which makes cost figures highly AUM-dependent. The problem arises when, in historical simulations, the underlying AUM used to size positions fluctuates as profits and losses accrue, drifting substantially over time.

If we want to study the net-of-fee feasibility of a signal at a precise reference AUM, yet the capital we actually deploy varies over time, the very premise of tying our estimates to a specific amount of capital falls apart.

To correct for this effect, in a 2025 paper titled The Tranching Dilemma, we introduced a cost-normalization procedure designed to keep cost figures comparable across drifting AUMs.

Specifically, transaction costs are always computed relative to a fixed reference AUM, and then rescaled so that their impact on the current (effective) AUM matches what it would be under the reference AUM.

This ensures that the effect of transaction costs (including both fees and market impact) remains consistent and comparable over time, independent of portfolio growth.

Our Case Study

To illustrate the above, we take as our case study an intraday mean-reversion signal that trades across a basket of liquid single-name equities, entering positions via Market-If-Touched (MIT) orders.

Often, these signals are simulated using canonical limit orders, which guarantee an execution price at or below our chosen level but do not guarantee that the full intended quantity gets filled. MIT orders exhibit the opposite behavior: they guarantee full execution once triggered but not price, so we should expect some slippage as the cost of transacting the entire intended size.

As always, the details of how the signal is constructed and how stocks are selected are provided at the end of the article.

To run our capacity simulation, we assume the following cost structure and apply the daily normalization procedure outlined above, across reference AUMs of $10K, $20K, $50K, $100K, $500K, and $1M:

• For transaction costs, we assume the standard Interactive Brokers tier of $0.0035 per share, with a minimum commission of $0.35 per trade (both values are doubled for sell transactions to conservatively account for SEC and FINRA regulatory fees).

• Market impact is estimated via I-Star model, with parameters a₁ = 708, a₂ = 0.55, a₃ = 0.71 (in our implementation, σ and ADV are computed using rolling windows that include the entire day of trade execution).

• We allow fractional-share trading, and the MOC orders sent to liquidate positions end-of-day are assumed to incur no slippage.

What becomes clear is that the erosion in risk-adjusted performance, much like the transaction costs driving it, scales non-linearly with traded size.

From a theoretical Sharpe ratio just above 1, introducing a realistically modeled impact of fees and slippage generates different erosion for different reference AUMs, suggesting that for this specific signal:

• The lower-bound solution of our capacity problem sits around a $20K reference AUM: below that level, the impact of fees becomes disproportionately large and erodes Sharpe too heavily.

• The ideal AUM is $50K, where net-of-fee Sharpe reaches its peak value.

• As the reference AUM grows, market impact becomes increasingly punishing: $100K appears to mark the upper bound of our capacity, beyond which trading would require more sophisticated execution algorithms to contain slippage.

What Can We Do?

Now that we have understood the outlines of the problem, we should ask ourselves whether there is anything we can do to “lift” the decay curve and improve risk-adjusted performance as our baseline AUM moves above or below the ideal value we identified.

On the left side of the erosion curve, the only way to ease the minimum-commission problem is to either find a broker that does not price fees with hard dollar minimums, or commit more capital to the strategy; otherwise, there is little room for intervention.

At the upper bound of the capacity issue, however, there is genuine room for improvement: as a matter of fact, our simulation assumes no sophisticated execution policy, and it sticks to plain-vanilla (“one-shot”) MIT orders.

What well-capitalized institutions and trading desks typically do is leverage their technological stack to build up desired exposures in ways that minimize market impact. In other words, participants who must execute sizeable quantities can ease into their positions through more numerous, smaller orders, usually within a maximum signal-to-execution time window.

Seen this way, execution becomes operationally more delicate as a function of how much we need to transact (Q) relative to the typical liquidity of the traded security (ADV) and the maximum allowable time (T) between signal generation and full execution.

Ideally, each “sliced” order remains large enough to avoid brokers’ hard commission minimums, yet small enough to keep slippage (i.e. market impact) in check, effectively lifting the right side of the Sharpe erosion curve.

The chart above illustrates what would happen if we could hypothetically slice each order into several (smaller) sub-orders while still securing the same execution price across them all. This order-slicing practice (destructive for smaller accounts) can meaningfully extend the signal’s estimated capacity at larger AUM levels.

For reference, through 4 sub-orders, the capital that maximizes net-of-fee Sharpe ratio climbs to $500,000: ten times the ideal AUM identified under one-shot execution.

Conclusion

All things considered, we hope to have made clear that the capacity dilemma is better framed as a three-way optimization problem, spanning how much capital a signal can, should, and must run on, and we hope the framework we’ve proposed can prove useful in your own work as a quantitative researcher or systematic trader.

Subscribe now

Since we’re currently planning a follow-up piece exploring more advanced intraday order execution techniques, we would recommend subscribing to ensure you are notified upon its release.

If you found this article useful, feel free to leave a comment and reach out via direct message or email at info@concretumgroup.com for any questions.

Strategy Rules & Methodology

The full set of trading rules used in the case study is summarized below, covering signal construction, the stock-selection universe, and the entry and exit logic implemented through MIT and MOC orders.

Read more

We also know that during downside moves, technical and behavioral forces can trigger positive feedback loops that exacerbate market sell-offs.

Investors cut risk, volatility-managed strategies deleverage, hedgers chase protection, and the price of downside insurance rises faster than fundamentals alone would seem to justify. In other words, downside moves are often made worse by the way investors react to them.

At the same time, speculators who have kept capital available for richer opportunities can step in, provide liquidity to forced sellers, and try to exploit abnormal corrections.

As quantitative traders, we often look to implied volatility to identify short-term panic regimes. The VIX, commonly referred to as the “fear gauge”, is the best-known measure of this kind: in technical terms, it reflects the level of implied volatility embedded in S&P 500 option prices, providing a market-based estimate of expected volatility over the next month.

In several of our previous studies, we have shown how investors tend to overpay for such protection. This creates profitable trading opportunities in VIX-linked instruments and derivatives…

One example is documented in our paper “The Volatility Edge”, which recently placed fifth in the 2026 Quantpedia Awards.

… but we think there is another way to exploit the VIX.

For instance, we can use it to identify when equity markets enter a temporary stress regime, when moves tend to become over-amplified by the technical and behavioral forces described above, and then implement the trade directly through SPY.

In this article, we present a simple system built around three core components:

1. A VIX-based signal to identify volatility bursts;

2. Position-sizing rules to exploit prolonged weakness;

3. Exit criteria designed to capture a potential market rebound.

From January 2007 through June 2026, our framework generated a 10.8% CAGR net of fees, while being active on only 14% of trading days.

The strategy does not require continuous intraday monitoring or discretionary intervention: signals are computed shortly before the close and implemented via market-on-close (MOC) orders, resulting in a relatively simple operational workflow.

In this article, we walk through how the signal is defined, how exposure is increased when stress persists, how the same-day implementation avoids look-ahead bias, and where the approach’s strengths and weaknesses appear in the data.

We also view this as the first step in a broader research project. The results are promising enough that, over the next few weeks, we plan to publish a follow-up piece applying the same methodology across multiple assets and implied-volatility indexes, along with Python code to help readers automate the strategy via Interactive Brokers APIs.

The VIX as a Signal

Read more

Which stocks are more likely to outperform?

At first, this is not a question about stories, forecasts, or narratives. It is a question about characteristics. Are cheaper stocks more attractive? Do recent winners keep winning? Do more profitable companies earn higher returns?

Once a pattern appears in the data, a second and deeper question follows: why does it exist?

One of the cleanest ways to study this problem is through decile portfolios. The idea is straightforward: at the end of each month, take a universe of stocks, sort them based on a specific metric, and divide them into ten equal groups (aka deciles). The first decile contains the stocks with the lowest values for that metric, while the tenth decile contains the stocks with the highest values.

Researchers then track the returns of each decile during the following month, repeat the procedure month after month, and study the resulting performance over time.

This simple framework has become one of the most widely used tools in quantitative finance because it helps researchers understand whether certain stock characteristics are associated with higher future returns.

The Value Effect

One of the most famous examples is the value effect. In this case, stocks are often ranked by their book-to-market ratio. Historically, companies with high book-to-market ratios, commonly referred to as value stocks, have tended to outperform companies with low book-to-market ratios, often called growth stocks.

This raises a fundamental question:

Are these return differences a fair compensation for bearing additional risk, or do they reflect market mispricing?

The traditional risk-based explanation, associated with Fama and French, is that value stocks are riskier. They often represent companies facing uncertainty, financial stress, or weaker growth prospects. Under this interpretation, their higher expected returns are compensation for bearing additional risk.

However, the behavioral finance perspective, advanced by researchers like Lakonishok, Shleifer, and Vishny, argues that investors systematically overpay for growth stocks and undervalue value stocks, leading to predictable mispricing. They attribute this behavior to representative bias, where investors naively extrapolate past earnings growth rates too far into the future.

Momentum: The “Premier Anomaly”

While the value factor can be understood as a compensation for additional risk, the momentum factor, introduced by Jagadeesh and Titman in 1993, challenged the efficient market hypothesis. Unlike value, which has a clear risk-based rationale, momentum relies purely on past price movements.

The classic momentum factor, constructed using historical stock prices, quickly earned the title of premier anomaly.

In momentum investing, stocks that have performed well over the past 12 months (excluding the most recent month) tend to continue outperforming, while underperformers continue lagging.

The chart below illustrates the annualized alphaof portfolios formed on past 12-month returns, demonstrating that the highest momentum stocks (Decile 10) have significantly outperformed the lowest momentum stocks (Decile 1).

However, long-short momentum strategies have historically suffered from sharp drawdowns, particularly during major market rebounds following large sell-offs (e.g., 1932, 1939, and 2009). These extreme reversals pose significant challenges for momentum investors, as past winners often experience abrupt underperformance.

A more refined approach, proposed by Pedro Barroso and Pedro Santa Clara, dynamically adjusts exposure to momentum based on recent volatility. Their risk-managed momentum strategy smooths the equity curve and mitigates extreme losses, making it a more resilient implementation of momentum investing.

The red line in the chart below illustrates this enhanced approach, demonstrating how adjusting for risk significantly improves the stability and long-term performance of momentum-based portfolios.

Why Does Momentum Exist?

Momentum is often explained through behavioral biases and limits to arbitrage. Investors may underreact to new information, allowing price trends to persist longer than they should. Later, they may overreact, pushing prices too far in one direction.

Our interpretation of trend persistence is closely related to the idea of slow-moving capital, a topic we discussed in a previous post.

In that piece, we argued that trends are often not the cause of market moves, but the visible trace left behind by a deeper adjustment process.

Even when information is public, capital does not always move instantly toward the most attractive opportunities. Investors face mandates, risk limits, career concerns, liquidity constraints, and operational frictions. A large institution cannot simply read a piece of news and immediately buy or sell billions of dollars of exposure without affecting the market. Information becomes price gradually because capital, interpretation, and execution all take time.

This slow adjustment can help explain why trends persist. At the same time, practical frictions prevent arbitrageurs from eliminating the anomaly completely. Trading costs, short-selling constraints, liquidity limits, and institutional restrictions all make it difficult to exploit these patterns perfectly.

Beyond Value and Momentum

Over time, researchers have identified many other characteristics associated with stock returns. Among the most studied are:

• Low volatility: stocks with lower historical volatility have often delivered attractive risk-adjusted returns.

• Profitability or quality: companies with strong earnings, stable cash flows, and high returns on capital tend to perform well.

• Investment: firms with disciplined capital expenditure policies have often outperformed more aggressive investors.

• Liquidity: less liquid stocks may offer a premium to compensate investors for higher trading costs and lower marketability.

However, the growing number of published factors has created a new problem: not every discovered factor is real. Some may simply be the result of data mining, where thousands of signals are tested until something appears statistically significant by chance.

How to Identify a Reliable Factor

To separate robust factors from statistical noise, researchers and practitioners often rely on a few common-sense criteria.

For example, Larry Swedroe and Andrew Berkin suggest that a factor expected to continue delivering excess returns should be:

• Persistent: it should work across different time periods and market regimes.

• Pervasive: it should appear across markets, geographies, or asset classes.

• Robust: it should survive reasonable changes in definitions and methodology.

• Intuitive: it should have a plausible economic or behavioral explanation.

• Investable: it should be implementable after accounting for transaction costs, liquidity, and real-world constraints.

These criteria do not guarantee that a factor will keep working in the future. But they help distinguish serious research from statistical coincidence.

Concretum Factor Tracker

One of the practical difficulties of factor investing is that monitoring these effects requires data, programming skills, and time.

In principle, investors can track dozens of characteristics and factor definitions, many of which also appear in the Kenneth French Data Library: size, book-to-market, operating profitability, investment, momentum, short-term reversal, long-term reversal, earnings-to-price, cashflow-to-price, dividend yield, accruals, market beta, and net share issues, among others.

To make this easier, we developed a free web application that provides up-to-date performance statistics and charts for a broad set of academic-style equity factors and stock characteristics.

All the charts used in this article were generated with this tool.

You can try this tool by clicking the button below!

Concretum Factor Tracker

The tool is currently FREE and designed for investors, researchers, professors, and students who want to explore factor investing in a more accessible way.

Since it is still evolving, feedback and suggestions are very welcome.

If you have any ideas or comments, feel free to contact us via email at info@concretumgroup.com or by leaving a comment below.

If you found this article useful, feel free to leave a comment and reach out via direct message or email at info@concretumgroup.com for any questions.

Disclaimer
This publication is provided by Concretum Group for informational, educational, and research purposes only. It does not constitute investment, financial, legal, or tax advice, nor a recommendation to buy or sell any security, instrument, strategy, or investment product. All investments involve risk, including possible loss of principal. Past performance, backtested performance, and historical analysis are not reliable indicators of future results. Readers should conduct their own research and consult qualified professionals before making investment decisions.

Full disclaimer: https://concretumgroup.com/disclaimer/

We wanted to know if it still works. So we ran it on the next 20 years of data, 2006 through March 2025, a period the strategy had no input in shaping. It does. Sharpe ratio of 0.68, CAGR of 6.05%, maximum drawdown of 11.7%, across the Global Financial Crisis, the COVID crash, and the 2022 inflation selloffs.

We published the full analysis in our paper Global Tactical Asset Allocation: Updated Results and Real-Market Implementation Using Python and IBKR, which is publicly available. We also built a Python notebook that automates the entire rebalancing process through Interactive Brokers.

Google Colab Notebook

This article covers the strategy in full: how it works, why a critical implementation detail can silently cost you more than 2% per year in CAGR, how to fix it, and how to run the whole thing automatically through a Jupyter notebook connected to a live broker account.

Subscribe now

What Is Tactical Asset Allocation?

Most long-term investors pick a fixed allocation and hold it through everything. 60% stocks, 40% bonds. They rebalance once a year and ignore the noise in between. That’s strategic asset allocation (SAA). It is simple, transparent, and easy to implement, but it also assumes that the optimal portfolio today is likely to remain the optimal portfolio through bull markets, bear markets, inflation shocks, and financial crises.

Tactical asset allocation (TAA) works differently. Instead of holding fixed weights through every market environment, it adjusts exposure according to a predefined set of rules. Different tactical approaches use different signals. Some rely on economic data, valuations, or market sentiment. Others, like the strategy examined in this article, use price trends.

The idea is simple. When an asset is in an uptrend, it is held in the portfolio. When the trend deteriorates, the position is reduced or eliminated, with the proceeds typically moved to cash or short-term bonds.

The primary goal is not necessarily to maximize raw returns, but rather to improve the balance between return and risk by reducing exposure during unfavorable market environments. Investors who lose 50% need a 100% gain just to recover. Cutting that peak-to-trough loss in half changes the long-term compounding math dramatically.

The key constraint is that this has to be systematic. No human judgment, no market calls, no emotion. A fixed rule that runs the same way every time, regardless of what the news says.

The Original Faber Strategy

The strategy allocates capital across five total return indices, each representing a major asset class with genuine long-term return potential.

The five assets:

The signal rule: at the end of each month, compare the closing price of each asset to its 200-day simple moving average.

• Price above the 200-day SMA: invest 20% in that asset

• Price below the 200-day SMA: put that 20% into 90-day T-Bills instead

No bottom fishing. No macro view. No optimization. Each asset gets an equal 20% weight when it’s in an uptrend, and gets replaced with cash when it isn’t.

Over the 1972 to 2005 in-sample period, this produced a Sharpe ratio of 0.81, a CAGR of 11.7%, and a maximum drawdown of 9.5%. Note that the drawdown figure uses monthly observations, so the actual portfolio drawdown was likely somewhat deeper.

The paper made a big impact. The strategy is elegant, transparent, and grounded in a well-understood behavioral phenomenon: momentum. Assets that have been going up tend to keep going up, until they don’t.

The question is whether those results were the product of a genuine edge or a well-fit backtest.

Updated Results: 20 Years Out-of-Sample

We extended the analysis from where Faber left off: January 2006 through March 2025. No in-sample tuning. No parameter adjustment. The same 200-day SMA, the same five assets, the same monthly rebalancing at month-end.

The period included some brutal stress tests:

• The Global Financial Crisis (2008-2009)

• The European Sovereign Debt Crisis (2011)

• The Taper Tantrum (2013)

• The China Crisis (2015)

• Volmageddon and the US-China Trade War (2018)

• The COVID-19 pandemic (2020)

• The inflation-driven selloffs of 2022

Despite all of it, the strategy held up.

Returns came in below the in-sample period, which is expected. Strategies rarely perform as well out-of-sample as they do on the data they were designed on. But a Sharpe of 0.68 with an 11.7% maximum drawdown, earned through two decades of genuinely difficult markets, is a strong result for something this simple.

The Hidden Problem: Rebalance Timing Luck

Here’s something most backtests don’t address.

The original strategy rebalances at month-end. But there is no particular reason why the last trading day of the month should be privileged over the 10th, the 15th, or the 20th trading day. These are all reasonable implementation choices for the same underlying strategy.

The question is not which day is “correct”. The question is how much the results depend on that choice. If two investors follow identical rules but rebalance on different days of the month, how different can their outcomes become?

To answer that question, we ran 21 separate simulations of the same Global Tactical Asset Allocation (GTAA) strategy. Each one rebalanced on a different trading day of the month: from n=0 (last trading day) all the way to n=20 (the 20th trading day). Same strategy, same signals, same assets. Only the rebalancing day differed.

The spread was striking.

Nearly 220 basis points of CAGR difference between the best and worst rebalancing day. That's not a rounding error. That's the difference between a strong result and a mediocre one, and it has nothing to do with skill or alpha. It's pure timing luck.

This is called rebalance timing luck. It is a well-known issue among asset managers and wealth management practitioners, yet it is rarely quantified in backtests. When signals are binary and rebalancing is infrequent, the exact moment you check the price relative to the moving average can determine whether you catch a trend early or miss it entirely.

The Fix: Tranched Weekly Rebalancing

The simplest solution is to rebalance more often. If you check signals weekly instead of monthly, you’re averaging across more observations, and the impact of any single unlucky day is smaller.

We ran weekly simulations on each day of the week. The CAGR spread shrank from 220 basis points to under 130 (and under 50 if you exclude Monday, which consistently underperformed).

But higher frequency rebalancing introduces a different problem: whipsaw.

When a signal flips from 1 to 0 and back again over several consecutive weeks, a weekly rebalancer will enter and exit the same position multiple times. A monthly rebalancer would have never moved. Each unnecessary round-trip costs transaction fees and can generate losses from adverse short-term price moves.

The faster you trade, the more exposed you are to noise.

To get the stability of monthly rebalancing and the timing benefits of weekly execution, we propose a tranched approach.

The idea: split the portfolio into four equal sub-portfolios, each rebalanced on a staggered four-week cycle. You still rebalance weekly, but only one quarter of the portfolio moves at a time. Week 1, tranche 1 rebalances. Week 2, tranche 2. And so on.

The practical impact:

The outcome dispersion drops to 63 basis points. Annual turnover falls by nearly half. At 10 basis points of transaction costs per trade, that turnover reduction alone saves around 35 basis points of CAGR per year. Not transformative, but real, and it compounds.

Signal Lag and Transaction Costs

Once the rebalancing structure is set, the next step is making the backtest honest about real-world friction.

We took the Wednesday-tranched portfolio and added two elements. First, a one-day execution lag: signals are computed from Tuesday’s closing prices, and trades execute at Wednesday’s close. This mirrors how the notebook works in practice. You download data, run the signal, prepare the order, and submit it the next day. Second, transaction costs of 10 basis points applied to every notional amount transacted.

Two findings worth noting.

The one-day lag slightly improved results. This runs counter to intuition but is what the data shows.

The transaction costs reduced performance by approximately 41 basis points per year. That’s a real cost, but it’s contained. SPY, EFA, IEF, DBC, and VNQ are among the most liquid ETFs in the world. The 10 bps assumption is conservative.

Automating It

The Python notebook handles the full rebalancing workflow: fetching data, generating signals, calculating trade sizes, and submitting orders to Interactive Brokers. It works for both the original monthly approach and the tranched weekly version.

The six ETFs used:

SHV serves as the cash proxy. Any capital not allocated to a trend-positive ETF gets invested in SHV. It always has a signal of 1 and can absorb any remaining allocation.

Step 1: Portfolio Setup

Before computing any signals, the notebook collects two inputs from the user through an interactive widget:

• Total AUM allocated to the GTAA strategy

• Current share counts for each of the six ETFs (only positions belonging to this strategy)

Step 2: Signal Generation

For each ETF except SHV, the notebook downloads one year of daily adjusted closing prices from Polygon.io and computes the 200-day SMA using yesterday's close as the most recent data point.

SMA_i = (1/200) x sum of P_i over the last 200 days

The signal for each ETF is then:

Signal = 1  if yesterday's close > SMA
Signal = 0  otherwise

ETFs with a signal of 1 receive a 20% weight. ETFs with a signal of 0 receive nothing. Adjusted prices are used throughout to avoid false signals from dividends and stock splits.

Step 3: Rebalancing Logic

With signals and weights defined, the notebook computes what the portfolio should look like and calculates the trades needed to get there.

Target shares for each ETF:

Target Shares = (AUM x Weight) / Last Price

For ETFs with a signal of 0, target shares are zero. The remaining unallocated capital goes into SHV.

Trade required per instrument:

Delta Shares = Target Shares - Current Shares

The output is a clean summary table:

Step 4: Trade Execution via IBKR

The notebook supports two execution paths.

Semi-automated via TWS Basket File. For investors who want to review orders before anything gets sent to the exchange, the notebook generates a .csv file containing the full order list. Each row specifies the instrument, action (Buy or Sell), quantity, and order type (Market-On-Close). You load this file into Trader Workstation using the BasketTrader tool and submit it manually.

Fully automated via IBKR API. For fully hands-off execution, the notebook connects directly to TWS or IB Gateway through the ib_async Python package, validates each contract, and submits Market-On-Close (MOC) orders programmatically.

MOC orders must reach the exchange by 3:45 PM ET to guarantee execution at the official closing price. The notebook generates and submits orders immediately after computing signals, leaving a comfortable buffer.

After submission, the notebook confirms each order with its ticker, quantity, direction, and order ID. Every instruction sent from Python is mirrored inside Trader Workstation, so you can visually verify that what the code produced matches what the broker received.

Running It in Practice

Start with paper trading. Connect to IBKR’s paper trading account (port 7497) and run a full rebalancing cycle before touching real money. Verify that the signals look right, that the target positions make sense for your AUM, and that the orders go through cleanly. Don’t skip this step.

Enter your current holdings accurately. The notebook takes your share counts as manual input. If those numbers don’t match your actual positions, the trade calculations will be wrong. Before every run, pull your live positions from IBKR and type them in carefully.

Don’t miss the MOC deadline. Market-on-Close orders have to reach the exchange by 3:45 PM ET. The notebook submits right after computing the signal, which is fast, but don’t introduce delays between opening the notebook and running the execution cell.

Think about tranche size relative to your account. Each of the four tranches represents about 25% of your total allocation. On smaller accounts, the individual order sizes within each tranche can get very small, especially after dividing 25% across five ETFs. If fractional shares aren’t supported for a given ETF, or if minimum commissions start eating into your orders, the monthly version is the cleaner choice.

Check the output every time you run it. Connections to IBKR can drop. Data feeds can go stale. Orders can be rejected if the contract validation fails or if the submission comes in too late. The notebook logs everything to a CSV, so the full audit trail is always there. Check it.

Conclusion

Meb Faber published this strategy in 2007 and the core idea hasn’t changed: trend-follow five asset classes with a 200-day moving average, move to cash when the trend breaks. Eighteen years later, it still produces a Sharpe ratio of 0.68 on data it never saw.

What our paper adds is two things. First, a real-world stress test: 20 years of out-of-sample performance through the worst market crises of the modern era. Second, an honest look at a problem most practitioners ignore: rebalancing on a single fixed day each month introduces up to 220 basis points of CAGR variance that has nothing to do with the strategy itself. Tranched weekly rebalancing brings that down to 63 basis points and cuts annual turnover by nearly half.

The Python notebook makes this implementable. Data from Polygon.io, signal generation using the 200-day SMA, trade sizing, and direct execution through Interactive Brokers via MOC orders. Semi-automated or fully automated, your choice. The notebook is publicly available on Google Colab.

Google Colab Notebook

Trend-following is not a free lunch. It underperforms during choppy, mean-reverting markets. It can stay in cash for extended periods and miss rallies. And running any automated system live requires ongoing attention to execution quality, data integrity, and broker connectivity.

Start on paper trading. Study the signals. Build your own conviction in the logic before you put capital behind it.

If you found this article useful, feel free to leave a comment and reach out via direct message or email at info@concretumgroup.com for any questions.

Disclaimer
This publication is provided by Concretum Group for informational, educational, and research purposes only. It does not constitute investment, financial, legal, or tax advice, nor a recommendation to buy or sell any security, instrument, strategy, or investment product. All investments involve risk, including possible loss of principal. Past performance, backtested performance, and historical analysis are not reliable indicators of future results. Readers should conduct their own research and consult qualified professionals before making investment decisions.

Full disclaimer: https://concretumgroup.com/disclaimer/

You can access the first two articles by clicking on the banners below.

As we continue to expand our book of strategies, we also try to present our readers with insightful pieces along the way. For obvious reasons we will not be able to disclose certain findings, but we do our best to make what we can available, and we are grateful for our readers’ support.

Continuing along these lines, we conclude the series by focusing on an aspect of trading baskets of single-name equities that we believe is often overlooked: the role of liquidity, not from a purely operational perspective, but as a potential driver of expected returns.

In particular, we investigate whether deeper or thinner liquidity conditions can materially influence the performance of a trading signal whose efficacy has already been established.

In other words: is there an illiquidity premium in short-term stock trading?

Liquidity Buckets

We start by taking the exact signal outlined in When Short Sellers Create Overnight Alpha, where we seek overnight exposure on a handful of selected stocks.

The most straightforward way to determine whether an illiquidity premium exists (i.e. whether we are compensated more for dealing in less liquid securities), is to apply the same signal on subsamples of our trading universe, with each subsample representing a different liquidity bucket.

In particular, at the beginning of each session we rank the names in our universe by their past liquidity and split them into five buckets (quintiles), with Q1 representing the most liquid stocks and Q5 the thinnest ones. The buy and sell engine remains untouched: what changes is just the pool of names it operates on (each bucket runs separately and independently).

Full details on stock selection and portfolio construction are provided at the end of this article, along with additional statistics for an alternative version of the signal that incorporates an operational filter.

Results

If an illiquidity premium existed, we would expect a roughly monotone increase in both raw and risk-adjusted performance metrics as we move from Q1 to Q5, as the signal is allowed to flag stocks in progressively less traded areas of the US market.

Judging from the AUM progression of the five strategies, the visual evidence for the presence of an illiquidity premium is already quite strong. For reference, some key gross-of-cost performance metrics are summarized in the following table.

It is striking to see the same signal perform with such a wide performance gap between the most and least liquid names in our universe. Between Q1 and Q5, CAGR increases more than tenfold and Sharpe jumps from 0.67 to a remarkable 3.15, with a maximum drawdown of only –16.5%.

Notably, the difference in performance does not appear to stem from different opportunity sets. All liquidity buckets deliver a comparable number of average entries per day: Q1 through Q4 sit tightly between 5.4 and 5.9, with Q5 running a busier portfolio at roughly 7.3 entries per day on average.

These numbers point firmly in the direction of a strong illiquidity premium at play: the thinner the traded names, the more attractive the expected returns (under the same signal).

The Liquidity Challenge

The idea of an illiquidity premium is not something we invented. It is a well-documented phenomenon in academia, where the underlying rationale is that investors should expect additional compensation for taking on the risk of transacting in illiquid securities.

Amihud and Mendelson (1986) originally linked expected returns to the width of bid-ask spreads, and Amihud (2002) later introduced a price-impact measure of illiquidity that explained variation in stock returns both cross-sectionally and in time-series.

Related to this is the finding that the small-cap factor can amplify other price anomalies, from broad factor-driven effects to signals like the one investigated in this series. Fama and French documented this for the asset-growth effect in Dissecting Anomalies (2008), and Asness et al. showed that once quality is held constant, a stable size premium emerges (Size Matters, If You Control Your Junk, 2018).

In a sense, when studied together, size and liquidity look less like standalone factors and more like amplifiers of whatever other effect a strategy is trying to capture.

So on one side our findings appear to confirm what the academic literature has long established… but that does not mean we should all rush to trade this signal, (or similar ones) in the most illiquid areas of the market.

The reason is simple: those returns, staggering on paper and gross of costs, are by construction extremely hard to harvest.

The table below shows the average liquidity recorded before entry and on the day of entry across all five liquidity buckets.

To put things in perspective, Q1 stocks typically trade more than $300M per day before entries flag, whereas Q5 stocks trade only a fraction of that, roughly $800K. For reference, numerous books and resources on stock trading recommend limiting entries to stocks that trade at least $1–5 million per day on average; by those standards, Q5 would be considered by many essentially untradable.

For example, in our 2025 paper Does Trend-Following Still Work on Stocks? we used $1 million (back adjusted for inflation) as a minimum liquidity filter.

It is true that, given the entry conditions we require, eligible stocks tend to experience elevated volumes on the day the signal fires. Viewed this way, Q5 becomes somewhat more accessible at around $11 million: that might give some hope, but the picture remains challenging.

Note - The volume information on entry day is available only at the end of the trading session, as the market closes, and gives us technically no time to execute. If we wanted to use liquidity at entry as a more precise proxy to filter our trading universe, we would need intraday data to backtest it rigorously.

Another issue worth flagging is that Q5 stocks are generally lower priced than Q1 stocks, with average entry prices being considerably higher in the most liquid bucket. This matters more than it might appear: typical broker commission pricing is structured as a fixed cost per share, therefore, lower-priced stocks will inevitably suffer a proportionally greater impact from fees, ultimately eroding a larger portion of the generated dollar P&L.

As the table shows, the average dollar P&L per share in Q5 is roughly half that of Q1, despite Q5’s vastly superior performance metrics, which can be traced back to the difference in average entry prices.

A final consideration is slippage. Although the signal we presented can be traded through MOC and MOO orders, both of which aim to minimise slippage by participating in the opening and closing auctions, attempting to place a meaningful order in a stock that trades only a few hundred thousand dollars per day will inevitably push the price against ourselves. This is, in effect, the same problem that the largest players in the industry experience in more liquid markets when deploying billions: the phenomenon is identical, but at drastically different scales.

Conclusion

Overall, we believe this experiment has provided some useful insight around a topic that may look unexciting on the surface but often uncovers fascinating dynamics of markets.

To keep it short, if you are a portfolio manager, or even a well-capitalized retail trader, your hopes of capturing a Sharpe ratio of 3 by trading illiquid names are a touch optimistic…

But if you are a small speculator, know that markets hide pockets of inefficiency and anomalies that your size of capital may be better suited to capture than that of larger participants.

It may seem counterintuitive, but in certain settings the small retail trader working with a few thousand dollars in their account is actually better positioned than those managing millions or tens of millions in allocation.

In a sense, it is Hector against Goliath. If you happen to be the small player, you should not feel discouraged. We suggest that the best thing you can do is put in the work and preparation to uncover the opportunities that are accessible to you and that larger players simply cannot get involved in.

When two investors run the same idea, how much of the difference in performance comes from their order size?

Sometimes knowing which games to play matters more than the strategy itself.

So… What Have we Learned?

Inspired by the upcoming edition of the Market Wizards book, we began this series by analyzing a promising short-selling signal in single-name equities, only to find it brought painful operational obstacles that would make its implementation quite impractical.

Rather than abandoning the idea, the very constraint of securing locates inspired us to explore long-sided opportunities, where we were able to find a more attractive variant of the original strategy… with solid foundations to explain and justify the returns we observed.

Today, we complete this three-part series by showing how something as seemingly unexciting as liquidity can carry important implications not only for execution, but for expected returns as well.

Hopefully, this three-part series has shed some light on aspects of quantitative research that are perhaps less fashionable and underrepresented through the lens of social media, yet constitute the backbone of our work on a daily basis.

If you have topics you would like to see covered in a similar format, feel free to reach out via direct message or email at info@concretumgroup.com. We are always happy to hear your suggestions.

Full Methodology and Extra Findings

In the following table, we describe step by step the rules we used to construct the study presented in this article.

Read more

One of the strangest facts in quantitative finance is that investors have historically earned a disproportionately large share of their long-term equity returns outside regular trading hours. In other words, much of the reward for owning stocks appears to have been generated between the close of one session and the open of the next.

As unusual as it may sound, this behavior happens to be one of the most widely documented stylized facts in quantitative finance, with evidence dating back to the influential work of Cooper, Cliff and Gulen (2008), who showed the effect to hold not only for individual stocks but for equity indices and index futures alike.

Seeing that many readers appreciated our last series of articles, in which we investigated overnight opportunities across stocks that are frequently targeted by active intraday short sellers, we thought it would be a good moment to present some additional findings around the overnight risk premium.

In particular, today we focus on some of the cross-sectional features that we find to be most predictive of positive overnight returns.

Is the Overnight Risk Premium Dead?

The first thing we do, is to assess whether market participants have indeed continued to be compensated for holding equity exposure overnight rather than intraday.

To do so, we construct two distinct equal-weighted benchmark portfolios that trade across a wide basket of Russell 3000 names. To ensure a minimum standard of liquidity, we restrict activity to the 2,000 most liquid securities within the index, where liquidity is measured as the median dollar volume exchanged on a rolling 252-day basis: we refer to this universe throughout as the Liquid Russell 3000.

The first benchmark portfolio simply buys every tradable stock at the open of day t and liquidates all positions at the close of the same session; the second buys at the close of day t, carries the positions overnight, and sells into the following open (t+1): importantly, both portfolios draw from the same subset of Russell 3000 names, with eligibility evaluated at the close of day t-1.

Even a visual inspection of the chart below reveals a clear divergence between the two portfolios. While the overnight strategy compounds steadily throughout the sample, the intraday portfolio struggles to generate meaningful growth, suggesting that a substantial portion of the equity risk premium has historically been earned outside regular trading hours.

For every calendar year in our sample, we calculate the Sharpe ratio of the overnight portfolio and compare it against the Sharpe ratio of the intraday portfolio: we focus on the difference between the two rather than on raw returns because the risk profile of overnight and intraday strategies is not constant through time.

Measuring excess performance on a risk-adjusted basis allows us to isolate whether the overnight window delivered superior compensation per unit of risk, rather than simply benefiting from periods of elevated market volatility.

Looking at a rolling three-year average of the spread, the overnight-minus-intraday dynamic appears to have been strongest in the years leading up to the publication of Cooper, Cliff and Gulen (2008). Since then, the effect has remained visible, although generally with less intensity.

In recent years (2022–June 2026), the effect appears somewhat weaker still, owing in large part to 2024 having been the only year in that stretch to deliver a strongly positive spread in the risk-adjusted returns of the two benchmark portfolios.

Judging from our universe, then, even though the overnight-minus-intraday dynamic may no longer be as persistent and clean as it was in the early 2000s, (something we might reasonably have expected) it remains clearly observable in recent years, and we would say it is far from dead or “arbitraged away”.

Predicting Overnight Returns

We now move to a more advanced step, one that goes beyond simply documenting the existence of the overnight risk premium.

Rather than searching for entirely new market anomalies, we ask a different question:

In other words, once a persistent market behavior has been identified, can we find features that help explain when it is likely to be more pronounced and when it is likely to be weaker?

We refer to these characteristics as satellite effects. A satellite effect is not a standalone anomaly in its own right, but rather a feature that enhances, amplifies, or helps explain the behavior of another anomaly.

In our case, the anomaly of interest is the overnight drift in equities: the objective is therefore to identify stock characteristics that may help predict where that drift is most likely to occur.

Read more

It won 5th place at the Quantpedia Awards 2026. The strategy compounds at 16.3% per year over a 17-year backtest, delivers a Sharpe ratio of 1, and keeps equity market correlation near 15%.

This article shows how to build an automated VIX volatility trading strategy using Python and Interactive Brokers. Based on our award-winning research, the strategy seeks to capture the volatility risk premium through systematic exposure to VIX-linked ETNs.

The paper is public. The strategy rules are fully transparent. And we built a Python notebook that automates the entire thing through Interactive Brokers.

This article is about the process of turning a research idea into a working automated trading system. We walk through the theory behind the strategy, the signal logic, the position sizing, and the code that connects it all to a live broker API.

This is meant as an educational exercise, not a production-ready system. The notebook is a starting point. We strongly encourage anyone interested to run their own tests, validate the signals on out-of-sample data, and stress-test the execution logic before deploying any real capital. The goal here is to show what the workflow looks like, not to hand you a finished product.

Let’s start with the theory.

Why Volatility Is Tradable

Most investors bet on direction. They buy stocks hoping prices go up, or they short hoping prices go down. Volatility traders don’t care about direction. They bet on how much prices will move, regardless of which way.

In practice, this means buying volatility when it’s cheap (usually when the crowd is calm) or selling volatility when participants are willing to overpay for it (usually when fear dominates and everyone wants portfolio insurance).

The key insight is simple: implied volatility, what the options market thinks will happen, almost always overestimates what actually happens. The VIX Index, which represents the market’s 30-day volatility forecast for the S&P 500, has overestimated realized volatility roughly 80% of the time over the past two decades.

That gap between what the market expects and what actually unfolds is called the Volatility Risk Premium (VRP).

Why does this premium exist? Because volatility tends to spike when equities fall. Portfolio managers buy protection through options and variance swaps, and the pool of risk-averse buyers persistently exceeds the pool of sellers willing to warehouse that risk. In equilibrium, sellers demand a premium. That premium is what we’re trying to capture.

From Variance Swaps to VIX ETNs

For decades, the only way to trade volatility directly was through variance swaps. These are OTC contracts between institutions, requiring ISDA agreements, large notionals, and counterparty infrastructure. Not exactly accessible to individual investors.

That changed in March 2004, when VIX futures launched on the Cboe Futures Exchange. And it changed again in January 2009, when Barclays launched VXX, the first exchange-traded note linked to VIX futures.

Suddenly, anyone with a brokerage account could trade volatility.

VXX tracks a rolling position in the two nearest-month VIX futures contracts, maintaining an average maturity of about 30 days. Think of it as a daily rolling bet on short-term VIX futures.

Today, several VIX-linked products exist:

• VXX / VIXY: Long volatility (1x exposure to 30-day VIX futures)

• SVXY: Short volatility (-0.5x inverse exposure)

• UVXY / UVIX: Leveraged long volatility (1.5x or 2x)

• SVIX: Short volatility (-1x inverse exposure)

The strategy in the paper uses SVXY to go short volatility and VXX to go long volatility. That’s it. Two instruments, one brokerage account.

The Dual Signal

The paper builds up to the final strategy through four progressive versions. Each one adds a layer of sophistication. We’ll skip straight to the logic that matters: the two signals that drive the final strategy.

Signal 1: Expected Volatility Risk Premium (eVRP)

This answers a simple question: is implied volatility currently higher or lower than what we expect realized volatility to be?

To estimate future realized volatility, we compute the standard deviation of the last 10 daily SPY returns and annualize it:

eRV30 = std(last 10 daily returns) * sqrt(252) * 100

Then we compare it to the VIX:

eVRP = VIX - eRV30

If eVRP is positive, the market is pricing in more volatility than recent history suggests. The premium looks attractive. If eVRP is negative or zero, the market might be underpricing risk, and shorting volatility becomes dangerous.

Signal 2: VIX Term Structure (Backwardation or Contango)

This answers a different question: what does the shape of the VIX curve tell us about market stress?

We compare the VIX (30-day implied volatility) to VIX3M (90-day implied volatility):

• VIX < VIX3M (Contango): The market expects calmer conditions in the near term. This is the normal state. Uncertainty naturally increases with time horizon.

• VIX > VIX3M (Backwardation): The market expects more volatility in the near term than in the medium term. This signals stress.

Combining these two signals gives us four regimes:

1. eVRP > 0 and VIX < VIX3M: Best case. The premium is attractive and the term structure is normal. Go short volatility at full size.

2. eVRP <= 0 and VIX < VIX3M: The premium has compressed, but the term structure is still calm. Stay short volatility, but at half size.

3. eVRP <= 0 and VIX > VIX3M: Both signals are negative. The expected premium is gone and the term structure is inverted. Flip to long volatility.

4. eVRP > 0 and VIX > VIX3M: Mixed signal. The premium looks attractive but the term structure says stress. Stay in cash.

Dynamic Sizing

Here’s where the strategy gets clever.

A static allocation to short volatility is dangerous because the risk of shorting volatility is not constant. When the VIX is at 12, even a modest shock can cause it to double. You collect small premiums while sitting in front of a steamroller. When the VIX is at 35, the odds of it doubling again are much lower, and the premium you’re getting paid is much richer.

The paper uses a dead-simple sizing rule: your allocation equals the VIX level divided by 100.

• VIX at 15? Allocate 15% of your portfolio.

• VIX at 30? Allocate 30%.

• VIX at 50? Allocate 50%.

This naturally reduces exposure during fragile low-VIX environments and increases it when volatility is elevated and the expected payoff is higher.

The full strategy rules are:

The “x 2” on SVXY is because SVXY only provides -0.5x exposure. To get the equivalent of a -1x short, you need twice the dollar amount.

The portfolio is rebalanced only when the current allocation drifts more than 2% from the target. This avoids unnecessary trading and keeps transaction costs low.

Subscribe now

The Results

Over the January 2008 to May 2025 backtest period, after 5 basis points of transaction costs per trade:

The strategy was short volatility on 90% of days, in cash for 6%, and long volatility for 4%.

What makes these numbers interesting is the combination: high absolute returns, near-zero correlation to equities, and a Sharpe of 1. Blending even a 10-20% allocation of this strategy into a passive SPY portfolio boosts the combined Sharpe ratio by roughly 20%.

Automating It

The strategy runs once per day, near the close.

The execution method is a Market-on-Close (MOC) order. A MOC order is submitted before the exchange cutoff (typically 3:50 PM ET on the NYSE) and gets executed at the official closing auction price. You don’t control the exact price, but you get the close, which is the most liquid print of the day. This is what institutional rebalancing flows use, and it’s the execution method assumed in the paper’s backtest.

Because MOC orders must be submitted before 3:50 PM, we need the signal ready before that deadline. The system computes signals at 3:45 PM ET. More precisely, it uses the closing price of the 1-minute bar that starts at 3:44 PM, which represents the market state at 3:45. This gives enough time to fetch data, compute the signal, and submit the MOC order before the cutoff, with a few minutes of buffer.

So the daily workflow is: at 3:45 PM ET, grab three numbers (VIX, VIX3M, and the 10-day realized volatility of SPY), compute the signal, determine the position size, and submit a MOC order. That’s it.

The full automation pipeline has four steps.

Step 1: Connect and Fetch Data

The system connects to IBKR’s Trader Workstation via the ib_async Python library. It pulls 30 days of daily SPY closes and one day of minute-level data for SPY, VIX, and VIX3M.

The last daily SPY close gets replaced with the 3:44 PM intraday price, so the realized volatility estimate includes today’s move up to the signal time.

Step 2: Compute the Signal

Using the data from Step 1:

1. Compute the 10-day standard deviation of SPY returns, annualize it

2. Subtract it from the current VIX to get eVRP

3. Compare VIX to VIX3M to determine contango or backwardation

4. Map the two signals to one of the four regimes

Step 3: Size and Place Orders

Once the signal is determined:

1. Compute the target allocation using VIX / 100

2. Convert that to a dollar amount based on account equity

3. Divide by the current price of SVXY or VXX to get share count

4. Compare the target weight against the current weight

5. If the difference exceeds the 2% rebalancing threshold, submit a MOC order

Step 4: Log and Disconnect

Every execution is logged to a CSV file with the full state: prices, signals, weights, target positions, orders placed, and timestamps. The system then disconnects from IBKR.

The Code

The full implementation is a Jupyter notebook with five cells. Here’s the structure:

Cell 0: Install packages. ib_async, pandas, numpy, pytz, ipywidgets.

Cell 1: Setup. Connects to IBKR, qualifies the five contracts (SPY, VIX, VIX3M, VXX, SVXY).

Cell 2: Data collection. Waits until the signal time, fetches daily and intraday data, extracts the 3:44 PM prices.

Cell 3: Signal and execution. Computes eVRP, checks term structure, determines sizing, fetches live prices for VXX/SVXY, calculates target shares, compares against current holdings, places MOC orders if the rebalancing threshold is breached, and verifies order acceptance.

Cell 4: Logging. Writes every field to a CSV and disconnects.

Cell 5: User interface. A widget-based form where you enter your account value, current positions, market close time, and rebalancing threshold. One click runs the entire pipeline.

The notebook is publicly available. You can find it in the Google Colab link referenced in the paper.

Google Colab Link

Subscribe now

Running It in Practice

A few things to keep in mind before you run this live.

Use paper trading first. Connect to IBKR’s paper trading account (port 7497) and run the strategy for a few weeks. Watch the signals, verify the orders, check the CSV log. Make sure the numbers make sense before you risk real capital.

Update your positions precisely. The current version takes your VXX and SVXY share counts as manual input. If you enter the wrong numbers, the orders will be wrong. Double-check before every execution.

Respect the MOC deadline. Market-on-Close orders must be submitted before 3:50 PM ET on the NYSE. The strategy computes signals at 3:44 PM and submits immediately after. That gives you a comfortable window, but don’t add unnecessary delays.

Watch for early-close days. A handful of trading sessions each year close at 1:00 PM ET instead of 4:00 PM. The notebook lets you adjust the market close time via a dropdown. On those days, select the 13:00 option so the signal time shifts to 12:44 PM and the MOC order goes out before the early cutoff.

This is not a set-and-forget system. The strategy is systematic, but the execution environment needs supervision. Connections drop, data goes stale, orders get rejected. If you want to learn more about what can go wrong, we wrote a separate article on Building Reliable Algo Trading Systems.

Conclusion

Volatility trading used to be the exclusive domain of institutional hedge funds with OTC infrastructure, ISDA agreements, and dedicated quant teams. That’s no longer the case.

With a standard brokerage account, two ETNs, and a few lines of Python, individual investors can now access the same volatility risk premium that institutions have harvested for decades. The strategy presented here is deliberately simple: two signals, dynamic sizing, and one daily trade. No options expertise required. No leverage. No complex hedging.

The full research paper is available publicly. The code is open source.

But simplicity does not mean safety. Volatility itself is volatile. Position sizing matters. Risk management matters. And the operational details of running live automated systems matter just as much as the strategy logic.

Use the paper as a starting point. Study the signals. Run the notebook on paper trading. Understand what you’re doing before you deploy capital. And if you need help building a more robust production setup around it, reach out.

If you found this article useful, feel free to leave a comment and reach out via direct message or email at info@concretumgroup.com for any questions.

Disclaimer
This publication is provided by Concretum Group for informational, educational, and research purposes only. It does not constitute investment, financial, legal, or tax advice, nor a recommendation to buy or sell any security, instrument, strategy, or investment product. All investments involve risk, including possible loss of principal. Past performance, backtested performance, and historical analysis are not reliable indicators of future results. Readers should conduct their own research and consult qualified professionals before making investment decisions.

Full disclaimer: https://concretumgroup.com/disclaimer/

For our readers, some methodological notes on how we construct the rebalance-fragility scores used to rank Nasdaq 100 names:

Read more