The rules
Hold the stock market from the start of November to the end of April, and Treasury bills (T-bills, short-term government debt) from the start of May to the end of October, switching at the month-end close (Bouman and Jacobsen 2002).
- Universe
- SPY and BIL
- Bars
- daily
- Position size
- 97% of equity in SPY from November to April and in BIL, a fund that holds T-bills, from May to October. Published: fully in the index, then fully in T-bills.
- Rebalance
- Twice a year, at the start of November and the start of May. Published: the same switches, at the month-end close.
What the research found
Bouman and Jacobsen (2002) studied total-return indexes (dividends reinvested) of 37 countries up to August 1998: 19 from 1970 (South Africa's only from 1973) and 18 only from 1988 or later (MSCI, except a Datastream index for South Africa). Winter (November-April) returns were higher than summer (May-October) returns in 36 of the 37, significantly in 20. They rule out risk and, except perhaps in the US, the January effect (high January returns). Only summer vacations survived their tests as a cause. In its October 2012 draft, Jacobsen and Zhang's study covered 108 markets over up to 319 years (55,425 monthly observations); the final version (revised 2018, published 2021) uses 62,962 observations. In that 2012 draft, winter returns exceeded summer returns by 4.52 percentage points (t = 9.69; a t-statistic above about 2 marks a gap unlikely to be chance, and the counts of significant markets here use a looser bar, about 1.65), and by 6.25 over the last 50 years; the published version puts the average gap at about 4%. In the same draft, out of sample (on later data), for Bouman and Jacobsen's 37 countries from September 1998 to mid-2011, the effect appeared in all 37, significantly in 15 by the authors' count. In the US the gap was 5.82 points in sample (t = 2.45) and 4.90 out of sample (t = 1.57, not significant). Trading the switch from October 1998 to April 2011 beat buy-and-hold in 31 countries on return and in 36 on the Sharpe ratio (return per unit of risk); in the US it returned 5.02% a year against 1.73% (Sharpe ratio 0.44 against 0.11), beating it in 46% of years. Jacobsen and Zhang leave out dividends, which Bouman and Jacobsen note biases results toward the strategy. Andrade and coauthors (2013) re-tested the same 37 markets with about 14 more years of data and found winter returns about 10 percentage points higher, significantly in 13. Their US test of this switch (SPY in November-April, one-month T-bills in May-October) averaged 4.64% a half-year over May 1994-April 2012 against 4.79% for holding SPY (Sharpe ratio 0.58 against 0.40). Maberly and Pierce (2004) attribute the US effect to October 1987 and August 1998, while Haggard and Witte (2010) find it robust to such outliers and the January effect, and better than buy-and-hold for its risk even after trading costs. Dichtl and Drobetz (2014) found that trading on it does not significantly beat the market.
Results
| Measure | Orbital test | SPY, same days |
|---|---|---|
| Window | Oct 31, 2016 to Oct 30, 2026 | Oct 31, 2016 to Oct 30, 2026 |
| Total return | +103.73% | +316.62% |
| CAGR | +7.38% | +15.35% |
| Sharpe | 0.28 | 0.65 |
| Discounted Sharpe | −0.02 | — |
| Max drawdown | −32.6% | −33.7% |
| Win rate | 73.7% | — |
| Profit factor | 3.79 | — |
| Trades | 19 | — |
Hypothetical: a backtest with modeled costs, not traded with real money.
Hypothetical growth of $10,000
R07 Halloween Effect · +103.7% vs SPY +316.6%
Rebuild it in Orbital
Paste this into Orbital Chat. The Strategy Developer builds the workflow and asks before it runs any backtest.
Build one program on daily bars that trades SPY and BIL; each buy uses 97% of equity, and every order fills at the next open. A program sees only each fund's own prices and the session date, with no ticker and no holiday calendar. So, on its first day, have it treat a fund as BIL if the average of (high - low) / close over that fund's last 20 daily bars, that day's included, is under 0.1%, and as SPY otherwise, and keep that for the whole backtest. Give it NYSE's full-day closures (holidays and one-off closures such as days of mourning) for every year of the backtest, so it can find the last trading day of April and of October. At the close of the last trading day of October, sell BIL and buy SPY; at the close of the last trading day of April, sell SPY and buy BIL. At the close of the backtest's first day, buy the fund for the half of the year the next trading day falls in: SPY for November to April, BIL for May to October. Never buy a fund that is already held or on order. Backtest the last 10 years with $100,000.
Is it real?
- Strategies tested together
- 7
- Beta to SPY
- 0.63
- Alpha vs SPY
- −3.5% a year
- Costs paid
- $2,845, 3% 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
Each switch fills at the open of the first November or first May session instead of at the month-end close. SPY therefore misses the move from October's last close to November's first open and keeps the move from April's last close to May's first open. BIL stands in for short-term T-bills: it is a fund that holds them, charges about 0.14% a year and pays its interest out as cash. Because a program cannot read its ticker, the two funds are told apart by their own daily price ranges. SPY stands in for each country's market index. 97% of equity, leaving room for costs. Four fills a year at 0.05% slippage (the assumed trading cost).
Sources