Global Seasonality Factor Investing

summary

The gist

Global Seasonality Factor Investing studies seasonality as a factor for predicting risk-adjusted returns across global futures markets and demonstrates that an equalweight ensemble of seasonality…

In short

The study tests five seasonality anomalies—day-of-week, month, quarter, half-year, and year—across global markets to see if they predict risk-adjusted returns. An equalweight combination of these models creates a factor portfolio that performs competitively with established factors like momentum and carry while being uncorrelated with them.

Key concepts

Seasonality Anomalies
These are recurring patterns in market returns tied to specific time periods, such as trading on Mondays or the end of a quarter. The paper investigates five specific types: day-of-week, day-of-month, day-of-quarter, day-of-half-year, and day-of-year.
Nadaraya–Watson Kernel Regression
This is a non-parametric statistical method used to fit prediction models to individual markets. It helps estimate seasonality by treating feature values (like the specific day of the year) as inputs to predict market returns, allowing for flexible modeling without assuming a strict mathematical form.
Seasonality Ensemble Advantage
Combining all five seasonality models equally into one portfolio significantly improves performance. This ensemble produces an out-of-sample Sharpe ratio that is competitive with existing global factors and is uncorrelated with them, adding about 30% to the net Sharpe ratio.

Terminology used across episodes

This episode discusses

The paper

Global Seasonality Factor Investing · Read on arXiv

Andrew Butler, Adam Butler, Â Maciej Zawadzki

ReSolve Asset Management

Transcript

Introduction to the show: ident: Quantitative Finance. The day in quantitative finance: models of markets, risk and prices.

Ruth: We're looking at the paper: "Global Seasonality Factor Investing".

Warren: Here's the gist of the paper. Global Seasonality Factor Investing studies seasonality as a factor for predicting risk-adjusted returns across global futures markets and demonstrates that an equalweight ensemble of seasonality models produces an out-of-sample Sharpe ratio that is competitive with, and uncorrelated to, established global time series factor portfolios The paper investigates five seasonality anomalies—day-of-week, day-of-month, day-of-quarter, day-of-half-year and day-of-year—to document their pervasive nature across bond, commodity and equity markets These factors produce excess returns not reasonably explained by established global time series carry, momentum and skewness factors The authors explore candidate economic rationales for the existence and persistence of each seasonality anomaly The market return forecast is cast as an empirical conditional estimation problem, where market returns are decomposed additively into drift, seasonality, and irreducible error To estimate the seasonality response variables, a non-parametric Nadaraya–Watson kernel regression model is applied to fit prediction models to individual markets based on feature values Feature variables are encoded as integer values representing the day within their sampling period, such as day-of-week or day-of-year Performance of Individual Seasonality Factors Empirical results show that fast-moving seasonal patterns, like day-of-week and day-of-month, produce strong gross performance but with extremely high turnover resulting in negative net performance Conversely, slower-moving patterns such as day-of-quarter and day-of-year produce positive gross performance and, with the exception of day-of-quarter, positive net performance The Seasonality Ensemble Advantage An equalweight ensemble of seasonality models produces an out-of-sample Sharpe ratio that is competitive with, and uncorrelated to, established global time series factor portfolios Adding seasonality as an equal-weight component to a global factor portfolio increases the out-of-sample Sharpe ratio by approximately 30% over the full evaluation period The ensemble produces aggregate turnover consistent with established time series factor portfolios The seasonality ensemble portfolio exhibits gross and net Sharpe ratios comparable to those of the carry, momentum and skewness factor portfolios The seasonality ensemble is nearly uncorrelated with all other factor portfolios and exhibits differentiated periods of performance Adding the seasonality ensemble factor to an equal-weight portfolio of carry, momentum and skewness improved the net Sharpe ratio by approximately 30% over the full out-of-sample period This establishes seasonality as a competitive global factor Appendix Details

Paper summary: Ruth: The paper "Global Seasonality Factor Investing" investigates five specific seasonality anomalies—day of week, day of month, day of quarter, day of half-year, and day of year—to document how pervasive these patterns are across bond, commodity, and equity markets.

Warren: They demonstrate that each individual seasonality factor produces excess returns that are not reasonably explained by established global time series factors like carry or momentum.

Iris: The methodology involves casting the market return forecast as an empirical conditional estimation problem where market returns are decomposed additively into drift, seasonality, and irreducible error.

Ruth: To estimate those seasonality response variables they apply a non-parametric Nadaraya–Watson kernel regression model to fit prediction models to individual markets based on feature values like day of week or day of year.

Warren: The empirical results show that fast-moving seasonal patterns, like day of week and day of month, produce strong gross performance but with extremely high turnover resulting in negative net performance.

Iris: Conversely, slower-moving patterns such as day of quarter and day of year produce positive gross performance and, with the exception of the day of quarter case, positive net performance.

Ruth: The authors then look at candidate economic rationales for the existence and persistence of each seasonality anomaly to explain why these cycles exist in different asset classes.

Warren: Ultimately, they show that an equal weight ensemble of these seasonality models produces an out-of-sample Sharpe ratio that is competitive with, and uncorrelated to, established global time series factor portfolios.

Iris: Adding seasonality as an equal weight component to a global factor portfolio increases the out-of-sample Sharpe ratio by approximately thirty percent over the full evaluation period <ref:2611.15657#pg1>.

Ruth: This means we’re looking at a statistically robust way to diversify risk using calendar patterns across global futures markets and assets.

Warren: The ensemble produces aggregate turnover consistent with established time series factor portfolios, so it doesn't introduce excessive transaction costs just because you added another layer of seasonality.

Conclusion: Ruth: So we’re wrapping up our look at "Global Seasonality Factor Investing." These authors argue that those day of the week, month patterns aren't just random noise; they are a factor you can actually use to predict risk-adjusted returns everywhere in the global markets.

Warren: Yeah, they’re showing how an equal weight mix of these seasonal models works well. They found this combination is competitive with established things like momentum and carry portfolios.

Iris: What's crucial here is that this new factor isn't just adding a little bit of noise to your portfolio; it’s providing returns that don't move with those other big factors you already use, like carry or skewness.

Ruth: That’s the main idea, right? It means you can diversify your risk by adding this seasonality piece without necessarily picking a completely new style of investing. They showed that adding this ensemble improves the risk-adjusted return measure by about thirty percent over a full testing period <ref:2611.15657#pg1>.

Warren: A thirty percent improvement in those returns over out-of-sample data is significant for anyone managing money. And they noted that the turnover in this ensemble stays consistent with what you'd expect from those traditional time series factor models, which means it doesn't introduce massive transaction costs just because you added another layer of seasonality.

Iris: It does sound compelling because it shows that this new factor isn't just adding noise; it’s providing an uncorrelated source of return. But what are the limitations they point out? Where does this approach fall short?

Ruth: They did mention that while the ensemble performs well, they are still exploring candidate economic reasons for each anomaly. That means the deep economic explanation for why these day-based cycles show up across all asset classes isn't fully settled yet. It’s a strong statistical finding but maybe not a complete theory on why those specific time patterns exist globally.

Warren: I agree with that. The paper shows seasonality is a real factor, but it doesn't give us the full picture of what the world is driving those specific day-based effects across all markets and asset classes yet. So what does this mean for building a portfolio?

Iris: It suggests that incorporating these seasonality factors into existing global factor portfolios can improve risk-adjusted returns by adding a source of return that is independent of carry, momentum, or skewness. This supports the idea that timing and cyclical behavior are valuable predictors in global markets.

Ruth: So for someone who only listens to the show, it means you could potentially add this seasonality component to your portfolio and see an improvement in your risk-adjusted returns by about thirty percent over a full out-of-sample period compared to just using the existing factors <ref:2611.15657#pg1>. That's what they are showing us.

Warren: It's worth noting that they demonstrated the ensemble is nearly uncorrelated with all other factor portfolios, which is a big deal when you’re trying to build a diversified risk strategy. This seasonality component provides something new to diversify against.

Iris: The paper does show that fast-moving patterns generate gross performance but negative net performance due to high turnover, while slower patterns are better for net returns except for the day of quarter case. That's a nice detail on how the timing impacts your trading costs versus your potential gains.

Ruth: It’s about understanding how different time scales impact your trading costs versus your potential gains. So, if you’re running a strategy, you need to know if you’re chasing super fast moves or just catching slower cycles.

Warren: And that feeds right into the question of turnover management. If the model suggests high turnover for day of week, but low for day of year, that tells you exactly where your trading activity should be focused.

Iris: The identification relies on this decomposition into drift, seasonality, and irreducible error to ensure they are actually measuring what they think they are measuring. It’s a conditional estimation problem.

Ruth: So the paper is establishing seasonality as a legitimate global factor for investing based on empirical evidence from five specific anomalies across multiple markets. It moves it from just an idea to a tested component of factor portfolios for risk management.

Warren: The key takeaway is that adding an equal-weight seasonality ensemble improves your risk-adjusted return measure by about thirty percent over out-of-sample periods and adds uncorrelated return sources <ref:2611.15657#pg1>. That's what we need to know right now.

More episodes

← Home