Secrets to Short-Term Trading Ch. 3: The Seduction of Calendar Effects

Investment Background

Larry Williams spent much of his career researching day-of-week and calendar-based patterns — end-of-month strength as pension flows arrive, pre-holiday drift, options-expiration-week effects — and found genuine, documented ones. Some calendar effects have real, defensible mechanisms behind them: end-of-month buying tied to actual, recurring institutional cash flows is not mysterious.

But calendar effects are also the single category of pattern most structurally vulnerable to looking real when it isn't, and this chapter's job is to explain exactly why, using the same discipline man-who-solved-the-market-simons ch03 applies to data integrity generally — here aimed at the specific, tempting case every short-term trader eventually runs into.

The Wall Street Translation

Why This Pattern Type Is Uniquely Dangerous

Here is the structural problem, stated precisely: a calendar effect divides your data into a small number of buckets — seven days of the week, twelve months, a handful of pre-holiday sessions per year — and with enough calendar buckets tested, some will show an impressive-looking historical edge purely by chance.

Test ten calendar-based hypotheses (day-of-week returns, month-of-year returns, pre-holiday effects, options-expiration-week effects, and so on) against fifteen years of data, and ordinary statistical variation guarantees that at least one or two will show a result that looks significant, even if none of them reflect anything real. The researcher who stops testing the moment a promising pattern appears — instead of also reporting the nine that showed nothing — has built an illusion without intending to.

The Confirmation Trap Layered on Top

The second layer of the problem: once a trader has found a calendar pattern that looks good in backtest, every subsequent live trade that confirms it feels like proof, and every trade that contradicts it feels like noise to be explained away — "the Fed announcement distorted it this month," "options expiration was unusual this cycle." This is the same selective-memory mechanism misbehaving ch3 describes for sunk costs, applied here to pattern confirmation instead of loss aversion.

A genuine calendar effect, if one exists, should hold up out-of-sample — tested on data that came after the pattern was identified, not just fitted to data that came before. Most calendar patterns that circulate in trading folklore have never survived this test; they were fitted to history, not discovered as a forward-looking regularity.

Division of Labor With the Rest of the Library

Book	Owns
man-who-solved-the-market-simons ch03	Data integrity as a discipline generally — lookahead bias, the danger of an exciting backtest result
misbehaving ch3	Selective confirmation and sunk-cost-style reasoning — the general psychological mechanism
This book	The specific, retail-scale version of both, applied to the single most common self-deception trap a short-term trader encounters: a calendar pattern that looks real because it was found by searching a small, convenient sample

Executable Trading Rules

1. Demand an out-of-sample test before trading any calendar pattern. Split the historical data in half by date. If the pattern only appears in the data used to discover it and vanishes in the later, untouched half, it was never real.

2. Count how many calendar hypotheses were tested to find the one that worked. If ten patterns were checked and one looked good, treat that one with proportionally more suspicion — the odds of a spurious "winner" rise directly with the number of things tested.

3. Prefer calendar effects with a stated, plausible mechanism over ones that are purely statistical. End-of-month pension-fund buying is a mechanism; "Tuesdays are historically strong" with no proposed reason is a pattern with no mechanism, and mechanism-free patterns decay faster once noticed.

4. Track a calendar pattern's live performance separately from its backtest, starting the day you begin trading it. The moment live results diverge meaningfully from the backtest, treat the backtest as the less trustworthy number, not the live result.

Relevance to a Retirement Portfolio

Calendar-effect trading itself has no place in a retirement core — it is short-term speculation, and even a genuine calendar edge is typically too small, after costs, to matter at retirement-portfolio scale.

What transfers is the general skepticism this chapter builds toward any small-sample, easily-searched pattern — including ones that show up in retirement-adjacent contexts, such as "this fund manager beat the market in seven of the last nine Januaries" or "this sector always does well in an election year." The same test applies: was this pattern found by searching many candidates until one looked good, and does it hold up on data collected after it was identified? A retirement investor who applies this same discipline to fund-selection folklore, not just to calendar-effect trading, avoids a much larger and more consequential version of the identical mistake.

Chapter 4 turns from data-driven self-deception to a purely behavioral trap: the cost of trading more often than a strategy's edge actually supports.