The Hypothesis
Translating a market intuition into testable, falsifiable predictions.
The original theory
"Did you notice that the stock market takes massive jumps on the days leading up to the influx of institutional money from 401(k)s, etc. for people that get paid on the 15th/end of month and every two-week cycles? Seems like collusion to drive up prices for regular people."
This is the claim we set out to test. Before we can run any statistics, we need to unpack the assumptions baked into it and turn them into concrete, measurable predictions.
Embedded premises
The theory contains four implicit claims, each of which must be true for the overall argument to hold:
- Predictable calendar: Semi-monthly (15th and end-of-month) and biweekly paydays create a known, repeating schedule of money flowing into the stock market via 401(k) contributions.
- Measurable price impact: The volume of money flowing in on these dates is large enough to move prices in a statistically detectable way.
- Anticipatory run-up: Prices rise before the money arrives, not just when it lands — implying that someone is buying ahead of the known flow.
- Intentional front-running: The anticipatory price movement is not random — it reflects deliberate action by institutional traders who know the calendar and trade ahead of it to profit at the expense of 401(k) participants.
Testable predictions
We translated the theory into five specific, falsifiable predictions. Each one maps directly to a statistical test:
| ID | Prediction | What we measure | "Supports theory" if... |
|---|---|---|---|
| P1 | Prices rise in the 5 trading days before paydays (T-5 to T-1) | Mean daily return in the run-up window vs. all other days | Run-up returns are significantly positive after controlling for known calendar effects (p < 0.05) |
| P2 | Prices rise on payday (T) and the 3 days after (T+1 to T+3) | Mean daily return on payday and post-payday days vs. all other days | Payday and post-payday returns are significantly positive |
| P3 | Volatility is elevated around paydays | Parkinson volatility in payday windows vs. non-payday days | Intraday range is wider around paydays than on ordinary days |
| P4 | Volume is elevated around paydays | Trading volume on payday-window days vs. non-payday days | Volume is significantly higher in the payday window |
| P5 | The price moves happen overnight, not during the trading session | Overnight return (prior close to open) vs. intraday return (open to close) | Overnight returns account for most of the excess return in the payday window |
What would DISprove the theory
A hypothesis is only useful if we can specify what evidence would cause us to reject it. The theory would be disproven if:
- No return difference: Average daily returns in the run-up window (T-5 to T-1) are statistically indistinguishable from returns on all other days, even after controlling for the turn-of-month effect and other known calendar anomalies.
- No volume or volatility signature: Trading volume and intraday volatility around paydays are no different from any other period.
- Pattern is constant across eras: If the pattern existed equally before and after 401(k) plans were created (pre-1980), it cannot be caused by 401(k) flows.
- No schedule sensitivity: If weekly, biweekly, and semi-monthly paydays all produce the same signal (or none at all), there is no evidence that the concentration of flow matters.
- Fails permutation test: If random date shuffles produce patterns as strong as the real payday calendar, the "effect" is likely noise. A Monte Carlo p-value above 0.05 would fail to reject the null hypothesis.
Possible alternate explanations
Even if we find a statistically significant pattern, it does not automatically mean the theory is correct. We must consider alternative explanations:
Turn-of-month effect
The turn-of-month effect is a well-documented anomaly (Ariel 1987, Lakonishok & Smidt 1988) where stock returns are disproportionately concentrated in the last trading day of each month and the first three days of the next month. Because end-of-month paydays overlap with this window, any payday signal could simply be the turn-of-month effect wearing a different label. Our regression controls explicitly for this.
No signal at all
The most parsimonious explanation is that there is no payday effect — that daily returns around paydays are indistinguishable from any other trading days once known calendar effects are removed. This is our null hypothesis, and the burden of proof is on the theory to disprove it.
Volatility compression
It is possible that what looks like a payday "bump" is actually reduced volatility around paydays creating an illusion of smoother upward drift, while the magnitude of returns is unchanged. Our GARCH analysis helps disentangle return effects from volatility effects.
Prediction scorecard: what actually happened
After completing the full investigation, here's how each prediction fared. The verdicts reflect ALL findings, including the Friday revelation and structural flow reframing.
| ID | Prediction | Verdict | What we found |
|---|---|---|---|
| P1 | Prices rise T-5..T-1 | REVISED | Yes, but only at lag+7-8 (not lag+0), only for semi-monthly pay, only on Fridays, and only post-1978. The effect is real but far more specific than predicted. |
| P2 | Rise at T and T+1..T+3 | NOT SUPPORTED | No consistent positive signal at T or T+1..T+3 after controlling for calendar effects and clearing lag. |
| P3 | Volatility elevated | REVERSED | Volatility is actually lower in the run-up window — consistent with "flow absorbed smoothly," not with manipulation. |
| P4 | Volume elevated | PARTIALLY | Volume is +10% on month-end paydays (the classic turn-of-month). Biweekly shows highly significant volume spikes (p<0.001) but NO return signal — the money is flowing but too diffuse to move prices. |
| P5 | Moves happen overnight | MIXED | Some post-month-end overnight return elevation (marginal), but not conclusive. The bigger finding is the intraday directional ratio: the market systematically closes near the high before month-end (p=0.001). |
How the theory evolved
Science doesn't often end where it started. Here's how each discovery reshaped our understanding:
April 16: "The pattern is at lag+7-8, not lag+0"
The naive test (lag+0) found nothing. Shifting by 7-8 trading days to account for the clearing lag revealed a statistically significant pattern. Theory revised: money doesn't impact the market on payday — it takes a week to arrive.
April 16: "It emerged with 401(k) adoption"
The clearing-lag pattern didn't exist before 1978 (when 401(k) plans were created). It emerged in the 1990s, peaked in the 2000s, faded in the 2010s. This timeline matches 401(k) adoption precisely — powerful evidence for the flow mechanism.
April 18: "The signal is 100% Friday-dependent"
Removing all Fridays from the dataset eliminates every significant result. The effect isn't about "lag+7-8" in general — it's about what happens on Fridays that fall in that window. Theory revised: the mechanism is Friday settlement clustering (payroll batching, biweekly overlap, options expiration interaction).
April 18: "Structural flow, not manipulation"
The combination of Friday dependency, biweekly volume-without-returns, component decomposition (neither 15th nor EOM alone is significant), and the Sabbatucci cross-sectional challenge points to structural settlement mechanics rather than intentional manipulation. The cost to investors is real but the cause is infrastructure, not adversarial intent.
- April 16: Original hypothesis stated. Clearing-lag discovery (lag+7-8). Historical emergence confirmed (post-1978). Alternative explanations tested (FOMC, OpEx, CPI — all survived).
- April 18: Friday isolation test — signal vanishes without Fridays. Biweekly deep-dive — volume without returns. Component decomposition — neither 15th nor EOM alone significant at lag+7-8. Settlement timing research — DOL regulations, recordkeeper data. Theory reframed from "manipulation" to "structural flow."