Initial Findings
The naive test results — before the clearing-lag discovery.
Executive verdict (before clearing-lag adjustment)
We tested whether S&P 500 returns are systematically higher around paydays, using the methodology described on the previous page. The short answer: the theory as stated is not supported.
When we align the payday calendar directly to market dates — treating the paycheck date as the date of market impact — we find no consistent evidence of elevated returns in the run-up window. In fact, the only coefficient that survives multiple-testing correction points in the wrong direction.
At this point in the investigation, the payday effect hypothesis appears to be dead in the water.
Prediction scorecard
In the hypothesis page, we translated the payday effect theory into five testable predictions (P1 through P5). Here is how each one fared against the data:
| ID | Prediction | Result | Verdict |
|---|---|---|---|
| P1 | Run-up returns (T-5 to T-1) should be positive and significant | Run-up coefficient is negative in most windows. Semi-monthly run-up = -0.001%/day (full period), with no significance. | Rejected |
| P2 | Post-payday returns (T+1 to T+3) should be flat or slightly positive from actual fund purchases | Post-payday coefficients are inconsistent across windows and pay types. No clear pattern emerges. | No support |
| P3 | Semi-monthly pay should show the strongest signal (concentrated flow from salaried workers) | Semi-monthly actually shows the weakest signal of all pay types tested. Mid-month shows the largest coefficient, but it's negative. | Reversed |
| P4 | The effect should be visible on payday (T=0) itself | There are some positive T=0 coefficients, but they appear on the wrong days and do not survive FDR correction. | Partial, wrong day |
| P5 | The effect should be stronger in recent decades (as 401(k) assets grew) | Some windows show larger magnitudes in recent data, but the signs are inconsistent — sometimes positive, sometimes negative. | Mixed |
Zero out of five predictions are cleanly confirmed. The scorecard reads like a failed experiment.
Coefficient estimates by bucket
The chart below shows the regression coefficients (in percent per day) for each timing bucket — run-up, payday, and post-payday — across all tested time windows and payday types, filtered to the S&P 500 with semi-monthly pay. Error bars show the 95% bootstrap confidence interval.
The coefficients above are daily excess returns — fractions of a percent. They look meaningless in isolation. But 401(k) investing isn't a single day. It's 24 purchases per year, for 30+ years, with dividends reinvesting on top. A 0.05% daily effect applied across that timeline compounds into real money — and we haven't even found the right day yet. The next page reveals where the effect actually lives, and the simulator shows what it costs over a career.
What we didn't know yet
At this point in the investigation, we were ready to file this under "debunked." But three major discoveries — each documented on subsequent pages — would change everything:
Discovery 1
The signal appears when you shift the calendar by 7-8 days to account for settlement time.
Discovery 2
That signal is 100% Friday-dependent — it vanishes on other days.
Discovery 3
Neither the 15th nor end-of-month alone produces the signal — it's a composite of both.
But wait...
Before we file this under "debunked," consider a fundamental assumption baked into every test above:
The theory assumed payday = market impact date. But your 401(k) money doesn't hit the market the day you get paid.
Think about what actually happens when you receive a paycheck with a 401(k) deduction:
- Your employer issues the paycheck (payday)
- The payroll provider batches 401(k) contributions and sends them to the plan recordkeeper (1-2 business days)
- The recordkeeper (Fidelity, Vanguard, etc.) processes the contribution and places trade orders (1-2 business days)
- Mutual fund shares are purchased at the next available NAV price (1 business day)
- The trade settles (T+1 for mutual funds, T+2 for equities)
That entire chain takes roughly 3 to 7 business days. So when we test "does the market go up before payday?" we're asking the wrong question. The market can't react to money that hasn't arrived yet.