Turn of the Month

Decades of academic and practitioner research have found that a large fraction of the equity market's long-run gains has historically accrued in a narrow window around month-end — the last few trading days of one month plus the first few of the next — rather than being spread evenly across all sessions.

This strategy monetizes that window with one fully mechanical trade per month:

• Buy SPY at the close of the 5th-to-last trading day of the month.
• Sell at the close of the 3rd trading day of the following month.
• Hold cash the rest of the time.

It is a market-timing strategy: entry and exit are fixed trading dates, not price levels. It recommends no individual stocks and defines no stop or target — its only risk control is the short, fixed holding period (about seven trading sessions) and being flat otherwise.

Best for

• Normal or rising markets where recurring, calendar-driven cash flows dominate short-term pricing.
• Periods of steady 401(k)/index-fund inflows and orderly month-end rebalancing.

Avoid when

• Sharp macro-driven selloffs and fast bear markets, when news overwhelms systematic flows (2002, 2008, 2022 all had negative TOM years).
• Environments where the effect has been arbitraged thinner — it is well known and has decayed since first documented.

Market behaviour exploited

Recurring, calendar-clustered flow, not news or fundamentals: month-end institutional rebalancing, payroll-driven retirement contributions, dividend reinvestment, and fund "window dressing."

Documented by Lakonishok & Smidt (1988) and revisited by McConnell & Xu (2008), who found that historically the entire market risk premium was earned in the turn-of-month window. It is a tendency, not a guarantee, and has weakened as it became widely known.

The rules are unchanged from the original strategy; a pure, date-driven core (turnOfMonthSignalFromBars) now re-expresses the SAME calendar anchors as an as-of signal so the backtest and the live strategy can never drift.

Entry

The 5th-to-last trading day of the month — nthLastTradingDay(year, month, 5).

Implementation: at the latest bar, entry fires on the first session on/after that anchor (previous bar before it, current bar on/after it). Because it keys off the trading calendar (weekends + NYSE holidays skipped), it lands on the correct session every month and fires exactly once. If the anchor day were an unscheduled closure, the next open session fires instead.

Exit

The 3rd trading day of the following month — nthTradingDay(nextYear, nextMonth, 3).

Implementation: the exit anchor is "the 3rd trading day of the month," detected the same way. A position opened on the prior month's 5th-last day is held until that session; the engine turns the standalone exit signal into a round-trip.

Position management

• One trade per month; flat (in cash) otherwise.
• No stop-loss, no profit target — the exit is a fixed date. entryPrice/stopLoss/firstTarget/riskPercent/rewardRiskRatio are reported as null (never invented).
• Holding period: 7 trading sessions in 398 of 401 backtested windows (three were holiday-shortened).

Deterministic, identical in construction to Santa Claus Rally and Russell Rebalancing — it uses the shared seasonalConfidence model. confidenceScore is a setup-quality measure (0–100), not a probability of profit.

• Inside the window → 100 (Exceptional).
• Approaching (≤10 days) → rises with proximity (e.g. 5 days out → 20).
• Far / finished → 0.

Bands (library-wide): <70 Weak · 70–79 Good · 80–89 Strong · 90–100 Exceptional.

The API returns ["Turn-of-month seasonal window is active"] while the window is Active, else []. Operationally:

• It is on/after the 5th-last trading day of the month.
• Position entered at that session's close in SPY.
• No stop or target set — the exit is the date, not a price.
• Exit at the close of the 3rd trading day of the next month.
• Flat (in cash) outside the window.
• Treat the historical edge as a tendency, not a guarantee — size accordingly.

• Structural, not predictive. The edge comes from recurring calendar-driven flow, not from forecasting news or fundamentals.
• Objective and fully mechanical. Entry/exit are unambiguous calendar rules — no discretion, no indicator tuning.
• Large, robust sample. 401 trades over 34 years and multiple regimes — far more statistical support than a once-a-year seasonal.
• Positive expectancy with a modest edge repeated often. 62.6% win rate, 1.85 profit factor; a small ~0.6% edge compounded across ~12 trades/year drove +873.9% over the test.
• Bounded exposure. In the market ~7 sessions/month, in cash the rest — the strategy sidesteps much of the between-window volatility.

• Small per-trade edge, decaying. ~0.6% per trade; well known and arbitraged, so future returns may compress.
• Real tail risk. Worst trade −9.99%, worst year −11.34%, 20.3% max drawdown — a macro shock landing in the window is taken in full (no stop).
• No adaptive risk control. A fixed-date exit means a bad turn-of-month is held to the exit date.
• Single-asset, single-window. No diversification within the strategy; frequent (monthly) trading makes it sensitive to commissions/slippage that the close-to-close backtest does not model.
• Regime dependence. The effect is strongest in calm/rising tapes and weakest exactly when it would help most (crises).

• **basis enum gap.** The historicalPattern.basis field is "published" or "derived"; there is no "empirical" value in the frozen v1 enum. The pattern is genuinely empirical (computed from our own completed backtest windows), so it is labelled "derived" with the real provenance spelled out in the note.
• Static snapshot. historical.* and the pattern are baked as of the 2026 windows; they drift as future months complete. The live /api/backtest always recomputes from current data.
• Close-to-close model. The engine fills at the signal bar's close and models no commissions or slippage — a real cost for a monthly-cadence strategy.
• Survivorship-free but single-vendor. Prices come from one source (Yahoo). SPY's history begins 1993-01-29, so the backtest starts there.