Alternative Explanations

Why this page matters most

Before claiming a finding is real, you must try to destroy it. This page documents every alternative explanation we tested — including some that genuinely challenge our conclusions.

Finding a statistical pattern is not the same as finding a cause. The pre-payday run-up could be an artifact of FOMC meetings, options expiration, inflation data releases, or the well-known turn-of-month effect. If the signal disappears once we control for these confounds, then the "payday effect" was never real — it was just a proxy for something else.

We tested three known calendar anomalies directly, engaged with two academic studies that challenge our hypothesis, confronted the turn-of-month overlap problem, and flagged several alternatives we could not test. What follows is the full accounting.

Known calendar anomalies

Three well-documented market events share calendar overlap with our payday run-up window. For each, we added a dummy variable to the regression and checked whether the payday signal (β_run) survived.

1. FOMC pre-announcement drift

What it is: The Federal Open Market Committee meets roughly eight times per year. A well-documented anomaly shows that equities rise +25 to +50 basis points in the 24 hours before the announcement (Lucca & Moench, 2015). Traders position ahead of expected rate decisions, creating a systematic upward drift before the meeting.

Why it could explain our signal: FOMC meetings are clustered in the middle and end of the month. If FOMC announcement days overlap with our lag+7/lag+8 settlement window, our signal could be FOMC drift in disguise — we would be measuring a well-known central-bank effect and calling it a payday effect.

The overlap: In the lag+7/lag+8 window, FOMC meeting days are actually under-represented — only about 40% of FOMC dates fall in the run-up window, compared to the ~57% you would expect by chance. The pattern is not being inflated by FOMC overlap; if anything, it is being diluted.

The controlled result:

Verdict: SIGNAL SURVIVES — the FOMC control has essentially zero impact on the payday coefficient.

2. Options expiration

What it is: Monthly options expiration occurs on the third Friday of every month — a date when massive derivatives positions are unwound, delta-hedging adjustments spike, and trading volume surges. This creates well-documented end-of-day price movements as market makers rebalance.

Why it could explain our signal: The third Friday frequently lands near the 15th of the month, creating mechanical overlap with semi-monthly pay dates. If you get paid on the 15th, lag+7 often lands near the 22nd — close to the third Friday. More importantly, our signal is 100% Friday-dependent, and OpEx is a Friday event. This is the closest mechanical confound.

The overlap: Approximately 20–25% of our lag+7/lag+8 dates coincide with the 3-day OpEx window (Thursday through the following Monday of expiration week).

The controlled result:

Verdict: SIGNAL SURVIVES — the OpEx dummy absorbs some variance but does not diminish the payday coefficient. However, see Section 2 below for why this verdict is more complicated than it appears.

3. CPI release dates

What it is: The Bureau of Labor Statistics releases the Consumer Price Index (CPI) — the main inflation gauge — typically between the 10th and 13th of each month. CPI surprises can move markets sharply: a hotter-than-expected print sends stocks down, a cooler print sends them up. These releases create directional moves that could contaminate any same-window analysis.

Why it could explain our signal: CPI releases fall between the 10th and 13th, landing near the middle of the month — close to where our 15th-of-month payday's lag+0 window sits. If CPI releases systematically create positive returns on those dates, we could be measuring an inflation-announcement effect.

The overlap: CPI release dates are modestly represented in the run-up window but with no systematic direction — CPI can push the market up or down depending on the surprise.

The controlled result:

Verdict: SIGNAL SURVIVES — CPI releases have no measurable effect on the payday coefficient.

Combined robustness check

The definitive test: add all three controls simultaneously. If the payday signal is really just a combination of FOMC drift, options-expiration effects, and CPI reactions, it should collapse when we throw everything in at once.

The Friday-OpEx tension

The controlled regressions above tell a clean story: add OpEx, signal survives. Case closed? Not quite. Our Friday deep-dive uncovered a finding that puts the OpEx verdict in tension with itself.

Here are the two findings side by side:

The problem: OpEx happens on the 3rd Friday. Our signal only exists on Fridays. A dummy variable flags specific OpEx dates, but if the signal is ONLY active on Fridays, the dummy might simply be saying "OpEx Fridays are no different from other Fridays" — while the real question is why Fridays matter at all.

The specific test we have not done: Is the payday signal stronger on 3rd Fridays (OpEx Fridays) compared to 1st, 2nd, 4th, and 5th Fridays of the month? If the answer is yes, OpEx is contributing to the effect. If no, then OpEx and the payday effect are genuinely independent Friday phenomena.

Until that test is run, the two findings coexist awkwardly:

• The dummy control says the payday signal is independent of OpEx
• The Friday dependency says the signal is concentrated on the same day-of-week where OpEx operates
• These are not contradictory in a strict statistical sense, but they leave an uncomfortable gap

Academic challenges

Two published studies directly engage with the "payday anomaly" hypothesis. We tested both.

3a. Ma & Pratt — "The Payday Anomaly"

Ma & Pratt (SSRN 3257064) studied day-of-month effects and found that the 16th of the month outperforms all other calendar days. Their hypothesis: semi-monthly paychecks typically arrive on the 15th, leading to 401(k) purchases on the 16th, which creates same-day buying pressure.

This is the closest academic parallel to our work. If Ma & Pratt are right, the effect should appear as a day-of-month phenomenon at day 16 — no clearing lag needed. We ran the same day-of-month regression on our 65-year dataset.

Our replication: NOT CONFIRMED

The 16th shows a negative coefficient in our data, and the p-value of 0.485 means we have essentially no confidence this is anything other than noise. The strongest day in our data is the 1st of the month — the classic turn-of-month effect.

3b. Sabbatucci — Stock Demand and 401(k) Plans

Sabbatucci (AEA 2024) asked a clever cross-sectional question: if 401(k) flow is what drives the turn-of-month effect, then stocks that are more heavily held by 401(k) plans should show a bigger turn-of-month effect. She sorted stocks by their 401(k) exposure and compared.

Her finding: High-401(k)-exposure stocks do NOT show a stronger turn-of-month effect. If anything, the effect is slightly weaker for high-exposure stocks — though the difference is not statistically significant.

This is a genuine challenge to the "401(k) flow causes the pattern" hypothesis. Three reasons it does not entirely invalidate our finding:

1. Different calendar anchor: Sabbatucci tested the classic turn-of-month window (last 1 + first 3 trading days). Our signal is at lag+7-8 from semi-monthly paydays — different calendar dates. Her null result applies to the traditional turn-of-month, not necessarily to the clearing-lag window.
2. Index fund mechanism: Most 401(k) equity investments go through S&P 500 index funds, which buy ALL constituent stocks proportionally. A cross-sectional test (comparing individual stocks by 401(k) exposure) would not detect an effect that hits all stocks equally through index purchases. The signal would appear in index-level returns but be invisible in stock-by-stock comparisons.
3. We do not claim exclusivity: We never claimed 401(k) flow is the SOLE cause of the turn-of-month effect. Multiple mechanisms — salary payments, dividend reinvestment, pension rebalancing — may contribute. Our hypothesis is that 401(k) flow is one contributor, not the only one.

But: even with those caveats, Sabbatucci's finding weakens the broad "401(k) flow drives market patterns" claim. If the mechanism were as strong as a simple flow-based story suggests, you would expect to see some cross-sectional variation — even with index funds in the mix, not all 401(k) money goes to the S&P 500. Target-date funds, sector funds, and active funds would create some differential exposure.

The turn-of-month question

Perhaps the most important "alternative" is the one our own data ultimately pointed to: the well-documented turn-of-month effect itself.

Our component decomposition showed three signals that may all be measuring the same thing from different angles:

The uncomfortable implication: the end-of-month signal at lag+0 IS the turn-of-month effect. The 15th's signal at lag+16-18 points to the SAME calendar dates from a different starting point. The combined semi-monthly signal at lag+7-8 is a composite that only reaches significance when both anchors are combined.

So is our "payday effect" just a different measurement of the turn-of-month?

Partially, yes. But the historical evidence provides an important distinction:

• The turn-of-month effect at lag+0 predates 401(k)s — it exists in our 1960s data, well before the Revenue Act of 1978 created 401(k) plans. This is the classic Ogden (1990) pattern driven by payment standardization.
• The clearing-lag component at lag+7-8 did NOT exist pre-1978 and emerged alongside 401(k) growth. In the 1960–1979 data, the lag+7-8 window shows no significant effect.

This means the clearing-lag component appears to be a NEWER, Friday-concentrated addition to the broader turn-of-month phenomenon — potentially representing the contribution of 401(k) settlement flow to a pre-existing calendar pattern.

Untested alternatives

Intellectual honesty requires acknowledging what we did not test. Four plausible alternative explanations remain untested, each for specific reasons:

Where this leaves us