Turn of the Month Effect Backtest: 33 Years of SPY, the Last Day and the First Three

00The verdict in three lines

• Does the turn of the month effect still work? Partly. Holding SPY for the last trading day of each month and the first three of the next made +0.273% a trade after costs over 404 months, but it passed four of our six gates, not six.
• It faded and came back: +0.420% a trade from 1993 to 2005, +0.086% from 2006 to 2015, +0.269% since 2016. Four days at random in the same months made a median of +0.142%.
• Buried, cause overfit by our rule (good first period, weak second). The surprise: the last trading day of the month lost on average; the first day was the best.

01The turn of the month effect

The turn of the month effect is the claim that US stocks earn an outsized share of their return in a few days around the end of each month. It is one of the oldest calendar effects in finance:

• Robert Ariel (Journal of Financial Economics, 1987) found that stock returns came mostly in the first half of trading months, counted from the last trading day of the previous month.
• Josef Lakonishok and Seymour Smidt (Review of Financial Studies, 1988), with 90 years of the Dow, located it in the last trading day of the month and the first three trading days of the next.
• John McConnell and Wei Xu (Financial Analysts Journal, 2008) found that from 1926 to 2005 the US market's average excess return outside that four-day window was about zero.

Why it could work: money arrives at the start of the month (salaries, pension and retirement-plan contributions, fund inflows), and some managers tidy their books at the month's end. Why it might not: it has been public since 1987, and published anomalies tend to shrink after publication (McLean and Pontiff, Journal of Finance, 2016). Our data start in 1993.

Our earlier tests on the same SPY file: the 200-day moving average and the Accumulation/Distribution line. This is the first calendar rule we have tested on stocks; we also compared night and day sessions on gold.

02How we made the rules mechanical

• Days: the trading days of the SPY file. Day -1 is the last trading day of a month; days +1, +2 and +3 are the first three trading days of the next. Holidays simply move the count.
• Turn of the month trading strategy: buy SPY at the close of day -2 (the second-to-last trading day) and sell at the close of day +3. The trade earns days -1, +1, +2 and +3: the window of Lakonishok and Smidt and of McConnell and Xu.
• Fills at the close (market-on-close orders). No signal is used: the dates are known from the calendar.
• Long only, all in, one trade a month, no stop. One variant, fixed before the run.

03How we tested it

SPY daily closes from its first day (29 January 1993) to 20 October 2026, 8,489 trading days, with all 136 dividends (Yahoo Finance history, the file of our 200-day test). Total return: each dividend reinvested at its ex-date close. Costs: 0.04% of the price a round trip, as in our other SPY tests, with 0.10% shown too. Unit: net % per trade. Periods by entry date: 1993 to 2005, 2006 to 2015, and a holdout from 2016 already used by our two earlier SPY tests. First-period bar t ≥ 2.0: one variant of a strategy not tested before on the site. Control: for every real trade, a four-day hold starting on a random trading day of the same month, 500 times. It keeps the number of trades, their length and their months, and removes only the position inside the month. Engine check: hand-made months (a holiday on the last day and on day +1, a dividend inside the hold, the incomplete months at the ends of the file) and an independent re-implementation that matched all 404 real trades and every neighbouring window (29 checks; tom_selftest_result.txt).

04Results: does the turn of the month strategy work?

Growth of $1, log scale. The strategy earns T-bill interest between trades. * Holdout already used before this test.

As a whole strategy, against buy-and-hold

Invested 19.1% of the days, the strategy made about half the yearly return of buy-and-hold with a third of the worst drawdown (measured at each close, open positions included). That is not proof of an edge: any rule out of the market four days in five has smaller crashes. Before costs it made +0.313% a trade; at 0.10% a round trip the sum of trades falls from +110.1% to +85.9%.

05Turn of the month or random days?

Four days anywhere in the same months made a median of +0.142%, because SPY rose on most days. The turn of the month made about twice that and beat 90.4% of the random sets, short of the 95% we require. And the rest of the month was not dead time: from the close of day +3 to the next entry, about 17 trading days, SPY made +0.608% a stretch (profit factor 1.50). Per day the turn of the month earned about twice as much, but unlike McConnell and Xu's 1926 to 2005 sample, the other days did not earn zero.

06Day by day: the last day of the month and the first

The first trading day of the month was the strongest day in our 33 years (+0.219%, up 62% of the time), and days -4 to -2 were positive too. But the last trading day of the month, which the classic window includes, averaged -0.041% and rose only 46% of the time. It is a descriptive finding, not a new rule: see section 09.

07By period and by year

* The holdout years were already used by our earlier SPY tests.

The strategy beat buy-and-hold in 11 of 34 years, mostly bear markets. Its best trade began on 30 March 2009, near the bottom of the financial crisis (+6.90%); its worst began on 30 July 2002 (-7.92%). Without the 10 best trades the sum stays positive (+56.7%), and without the 20 best too (+14.5%).

08Quarter-end months

Turns after a quarter's end did better than the others, as the window-dressing story would suggest, but with 135 trades the difference is well within chance; it is descriptive only.

09Other windows (descriptive, looked at after the gates)

Every window was positive. Starting earlier, before day -1, helped: days -3 to +3 passed all six gates and days -2 to +3 passed five. We saw that only after the results, so neither counts. Picking the best window after seeing the data is exactly what our protocol forbids, and with five windows looked at, one of them would look good by chance. A window from day -3 would need new data, or a test in real time, before it means anything.

10The gates

Overfit, in our rule, means it passed the first period and failed the second. Nothing was tuned here, since the window comes from papers written before our data; the label says the edge was not stable enough. There are no gaps in the file other than real market closures, so the data-gap check does not apply.

TWhat this shows, and what to take into your own tests

What the results show

• Holding SPY from the close of the second-to-last trading day of each month to the close of the third trading day of the next made +0.273% a trade after costs over 404 months, 1993 to 2026, with a profit factor of 1.41.
• It failed two of our six gates: the 2006 to 2015 period had a profit factor of 1.11, and four days at random places in the same months made a median of +0.142%, so the turn of the month beat only 90.4% of the random sets.
• The best day of the month was the first one (+0.219% on average, up 62% of the time). The last trading day, which the classic window includes, averaged -0.041%.

What remains unknown

• Whether the effect comes back or keeps fading after October 2026: our holdout years had already been used by two earlier SPY tests.
• Other windows, such as days -3 to +3, which looked better after the run. They need new data before they can count.
• Other markets and futures with leverage, and taxes on twelve round trips a year.

Why it was buried

The effect was strong from 1993 to 2005, weak from 2006 to 2015 and partly back after 2016. By our rule the cause is overfit: good in the first period, not in the second. Here that does not mean we tuned anything, since the window comes from papers written before our data; it means the edge was not steady enough. And the market rose on most other days too, so four days at the turn of the month were only somewhat better than four days chosen at random.

What you can apply to your own tests

• Compare a calendar rule with the same number of days at random in the same months: being invested at all already earns the market's drift.
• Split the result into periods. An effect that works in one decade and not the next is not something to bet the whole account on.
• Look at each day of a window separately; a famous window can carry a day that hurts it.

A note on the numbers: drawdown figures are measured on closed trades unless a table says it includes open losses. Open losses inside a trade are not counted in a closed-trade drawdown, so real account risk is higher, especially for systems with no stop. Backtests are not live trading.

FAQCommon questions

What is the turn of the month effect?

The finding that US stocks earn an outsized share of their return around the end of each month, usually from the last trading day of a month to the third trading day of the next (Lakonishok and Smidt, 1988; McConnell and Xu, 2008).

Does the turn of the month effect still work?

Partly, in our test of SPY: +0.42% a trade from 1993 to 2005, +0.09% from 2006 to 2015 and +0.27% since 2016, against a median of +0.14% for four random days in the same months. Buried, four of six gates.

Is the last day of the month good for stocks?

Not in our 33 years of SPY: the last trading day averaged -0.04% and rose only 46% of the time. The first trading day of the month was the strongest, at +0.22% and up 62% of the time.

Is a turn of the month strategy better than buy-and-hold?

Not in return: with T-bills between trades it made 5.2% a year against 10.8% for buy-and-hold, while invested about a fifth of the time. Its worst drawdown, measured at each close with open positions included, was much smaller (-18.0% against -55.2%), as for any rule mostly out of the market.

How was the turn of the month effect tested?

On SPY daily closes with dividends, 1993 to 2026, with the window fixed before any number was computed, 0.04% of costs per round trip, and rules written and dated before the data was run, costs included on every trade, results split into development, validation and holdout periods, a random-entry control as the benchmark, and every variant reported. Read the protocol.

11Check our arithmetic

Code provenance: our own code, written from the specification.

12Data notes

Source. Yahoo Finance daily history of SPY (SPY has never split; dividends listed separately), read from the saved history page; our close-plus-dividends index agrees with Yahoo's adjusted close to 0.05% over 33 years. Gaps. Only real market closures (September 2001, 2 January 2007, October 2012); the day count simply skips them. Fills at the official close; real market-on-close orders can differ slightly. Not included: taxes; SPY's yearly fee is already inside its price. Exploratory: one ETF, one window, a holdout already used.

13Corrections

4 Nov 2026, correction: the first name of Robert Ariel, the author of the 1987 study, was given as Richard. No result changed.

+More tests and guides

• 200 day moving average strategy
• Accumulation distribution line stocks
• Best time to trade gold
• Internal bar strength strategy
• Free prop firm challenge calculator
• What all the backtests found so far
• The backtesting protocol
• How to backtest a trading strategy without fooling yourself
• Backtesting glossary