Implemented Strategies: Calendar Anomalies Part 3

[WITH CODE] Creating and testing a calendar-anomaly strategy against the S&P 500

Reviewed and updated 20 July 2026

Hello!

Today, I will be revisiting an aggregate calendar-effect strategy that I created from the results in my prior two calendar-effect posts (Part 1 and Part 2). The selected strategy produces higher no-cost ending wealth and risk-adjusted returns than S&P 500 price-index buy and hold in this sample. However, the result is post-selection, and the advantage does not survive modest trading costs.

In my Recent Academic Research posts, I’ve highlighted papers examining reported return patterns in the U.S. stock market. The earlier posts in this series tested whether those patterns appeared in the historical sample.

This is the final post of this three part series. I enjoyed creating a multi-part series on this paper, as I was able to dive deeper into the calendar effects (and test how the calendar effects interact in an aggregate strategy).

Let’s get into it.

---

Paper Introduction

In the Recent Academic Research post on January 25, 2025, I featured a paper by Hussein Mohamed on index calendar anomalies. I have attached my summary of the paper below, and the full paper can be found here.

Summary of “Time-Based Trading Patterns”

This paper examines time-based trading patterns across major stock indices. Its results vary by index and by anomaly. For the S&P 500 specifically, the regression reports significant positive coefficients for the Halloween, Turn-of-the-Month, options-expiration, and Friday effects; a significant negative coefficient for the pre-holiday effect; and a weaker negative Monday coefficient at the 10% level. The FOMC, sports, January, and September coefficients are not statistically significant for the S&P 500.

For the S&P 500, the paper studies 1958 through the end of 2023 and trims the highest and lowest 2.5% of daily returns before estimating the regression. The S&P 500 interaction between FOMC meetings and the Halloween period is also not statistically significant. The backtests below ask separate trading-rule questions; they are not direct reproductions of the paper’s regression definitions.

I created two selected strategies, one for total return and one for Sharpe ratio. They were nearly identical in their included calendar effects. I keep those selections fixed below rather than searching for a new combination.

Below, you’ll find the S&P 500 price-index metrics for the effective backtest period, 4 January 2000 to 30 December 2024. Dividends and cash interest are excluded.

S&P 500 Buy and Hold Metrics (2000 to 2024)

• Ending Wealth of $1: 4.06

• CAGR: 5.77%

• Annualized Mean Return: 7.50%

• Annualized Standard Deviation: 19.39%

• Sharpe Ratio: 0.39

---

Recap of Prior Calendar Anomalies

Over the past two posts, I’ve tested several well-known calendar anomalies from academic research, comparing their performance against the S&P 500 from 2000 to 2024. Some strategies delivered strong risk-adjusted returns, while others completely failed to hold up in historical backtests.

• Turn of the Month — The rule invests on the final exchange session and the first three exchange sessions of each month. Its comparable full-period Sharpe ratio is 0.30. The reference input ends on 30 December 2024, one session before the official year-end close, so that return is retained but is not labelled month-end.

• Months and weekdays — November (0.38), July (0.33), and Thursday (0.24) have positive comparable Sharpes. A short September rule has a comparable Sharpe of 0.26.

• Scheduled FOMC statement days — The rule uses 199 scheduled statement dates from the Federal Reserve calendar, known before the prior close, and excludes unscheduled actions. Its comparable Sharpe is 0.56. It does not establish a dependable crisis hedge.

• Options expiration — The short rule uses the third Friday, moved to the preceding exchange session when that Friday is closed. Its comparable Sharpe is 0.20 before costs and short-borrow assumptions.

• Exchange closures — The long rule uses the return ending on the final session before a closure, entered at the preceding session’s close. It includes scheduled holidays and the 2004 one-off closure documented before entry. The 2007 and 2018 mourning closures, 9/11 shutdown, and Hurricane Sandy shutdown are excluded because the available evidence is too late or insufficiently timed for entry. Its comparable Sharpe is 0.37. A closed holiday itself has no tradeable close-to-close return.

• Super Bowl window — The short rule uses the seven observed trading sessions before each game. Its comparable Sharpe is 0.18.

These are descriptive historical associations. They do not establish statistical significance after the number of rules examined, and they do not establish implementable trading performance.

Now, let’s dive into Part 3.