Turn of the Month

The rules

Hold the stock index over the turn of each month and Treasury bills (T-bills, short-term US government debt) otherwise. Lakonishok and Smidt (1988) and McConnell and Xu (2008) define the turn as the last trading day of a month and the first three of the next; Hensel and Ziemba (1996), who tested this switch on the S&P 500, used the last trading day and the first four. Etula, Rinne, Suominen and Vaittinen (2020) find the gains from three trading days before the last one through three after it.

Universe

SPY

Bars

daily

Position size

97% of equity in SPY during each window, cash otherwise. Published: fully in the index during the window, T-bills otherwise.

Rebalance

Month-end, about seven trading days a month. Published: the last trading day and the first three of the next (Hensel and Ziemba: the first four; Etula et al.: three before the last day through three after).

What the research found

Lakonishok and Smidt (1988), with Dow Jones data for 1897-1986, and Ariel (1987), with US index data for 1963-1981, found stock returns concentrated around the turn of the month. McConnell and Xu (2008), using US market indexes, found that over 1987-2005 the value-weighted index (weighted by company size) averaged 0.15% a day over the last trading day and the first three of the next month, against -0.001% on the other days; over 1926-2005 it averaged 0.16% against 0.01% (0.23% against 0.05% equal-weighted). The gap held in both halves of 1987-2005, contrary to the finding of Maberly and Waggoner (2000, cited by McConnell and Xu) that it disappeared from S&P 500 futures (contracts on the index) after 1990. McConnell and Xu find the effect in 31 of 35 countries, and cite Hensel and Ziemba (1996), whose switch between the S&P 500 and T-bills beat buying and holding by 0.63 percentage points a year over 1928-1993. Kunkel and Compton (1998) report that this switch held the index over the last trading day and the first four of the next month, and ignored trading costs and taxes. In Etula and coauthors' (2020) data, July 1995 to December 2013, the S&P 500 averaged 0.11% a day on the three days before the last trading day and 0.13% on the first three of the new month, against 0.04% on the average day. Etula and coauthors tie the pattern to institutions raising cash for month-end, timed by settlement (the days a sale takes to pay), and say it seems to have grown more pronounced over time, as in 2003-2013. Their data predate the US moves to two-day (2017) and one-day (2024) settlement; when much of Europe went to two days in October 2014, the pattern moved toward month-end. Against this, Plastun and coauthors (2019) find the effect faded on the Dow in the 1980s and was gone by the 2000s: in 2010-2018 the turn days averaged 0.03% a day against 0.05% for the rest of the month, and their simulated trades were not significantly profitable after 2000. All of these figures are before trading costs. By Orbital's arithmetic on McConnell and Xu's figures, the four turn days carried all of the value-weighted index's return in 1987-2005 and about four-fifths of it in 1926-2005, when the other days' 0.01% a day was too small to tell from zero; on Etula's, the seven days around month-end carried most of the S&P 500's average return from July 1995 to December 2013, but not all of it.

Results

Hypothetical: a backtest with modeled costs, not traded with real money.

Hypothetical growth of $10,000

Rebuild it in Orbital

Paste this into Orbital Chat. The Strategy Developer builds the workflow and asks before it runs any backtest.

Build a program on daily bars for SPY. A program sees only its own prices and the session date, with no holiday calendar, so give it NYSE's full-day closures (holidays and one-off closures such as days of mourning) for every year of the backtest, and count only days the market is open; half-day sessions count. At the close of each month's fifth-to-last trading day (four trading days left), buy SPY with 97% of equity at the next open. At the close of the third trading day of the next month, sell it at the next open; if that day has no bar, decide at the close of the next day that has one and sell at the following open. Buy only on a fifth-to-last trading day, so a backtest that starts inside a window stays in cash until the next one. Stay in cash otherwise. Backtest the last 10 years with $100,000.

Is it real?

Strategies tested together

7

Beta to SPY

0.31

Alpha vs SPY

−2.8% a year

Costs paid

$14,910, 21% of the gain before costs

Data

Daily bars, 2,514 sessions

Fills

Next open at market, sized at the fill

Read this before you trust it

SPY is bought at the open of each month's fourth-to-last trading day and sold at the open of the fourth trading day of the next month. That is Etula and coauthors' window (three trading days before the last one through three after it) shifted by half a day: orders fill at the next open, so the position misses the night before the window and holds one extra night after it. The window is the same in every year of the test, although US settlement moved from three days to two in September 2017 and to one in May 2024. The classic window of Lakonishok and Smidt and of McConnell and Xu is narrower: the last trading day and the first three of the next month. Cash earns nothing between windows, where the published comparisons hold T-bills. SPY stands in for the studies' indexes: broad US market indexes, the Dow and the S&P 500. SPY's quarterly dividends go to whoever holds it in mid-March, June, September and December, outside every window, so the strategy collects none of them. 97% of equity, leaving room for costs. About 12 buys and 12 sales a year, each charged 0.05% slippage (the assumed trading cost): roughly 1.2% a year.

Sources