A t-statistic of -12 and Still No Trade: Why Real Effects Don't Pay
Daily t-statistics of -8 to -18 confirm short-horizon mean reversion exists. But decile spreads of one to four basis points sit inside a four-basis-point round trip.
By Daniel Okafor
3 min read

What's News
- A TradingView researcher ran a three-day logger in September storing one-minute bars for equities and crypto pairs, with roughly a year of history backfilled.
- Over 273 days of equity data, daily t-statistics on the reversion effect ran between -8 and -12; over 362 days of crypto, between -13 and -18.
- Decile spreads between 'cheap' and 'dear' measured one to four basis points against a costed round trip of four basis points.
- Ninety-six pre-declared tests in three families, Bonferroni-corrected, produced zero passes; a frozen holdout was never opened.
- The write-up concludes the effect is a liquidity provider's reward, not a trading signal for a taker.
Daily t-statistics between -8 and -12 confirmed short-horizon mean reversion in equities — and the effect still earned nothing for the trader who measured it. That is the core finding of a research write-up published on TradingView by user harrell.jordon, run on the State Street SPDR S&P 500 ETF (BATS:SPY) and a basket of crypto pairs. The statistics are decisive. The economics are not.
The experiment was narrow by design. For three days in September, the author ran a logger that stored every one-minute bar for a basket of equities and crypto pairs, with about a year of history backfilled behind it. The question was split in two: does short-horizon mean reversion exist in this data, and can a taker — a trader who crosses the spread — collect it?
How strong is the effect?
The first answer is yes, and the author states it plainly: "it is not close." Over 273 days of equity data, the daily t-statistics on the reversion effect ran between -8 and -12. Over 362 days of crypto data, they ran between -13 and -18. For scale, most academic finance treats a t-statistic near -2 as evidence an effect exists. These numbers are several times that threshold.
Anyone who has wondered whether short-horizon mean reversion is real now has an answer in this dataset. The effect is present in both asset classes, on thousands of one-minute bars, over hundreds of trading days.
Why can't a taker collect it?
The second answer ends the trading case. The decile spreads — the gap between what the measure flags as cheap and what it flags as dear — came in at one to four basis points. The round trip the author costed against was four basis points. The whole effect fits inside the fee.
This is where the statistical case and the economic case diverge. A t-statistic of -12 proves an effect exists. It says nothing about whether the effect is larger than the cost of capturing it. As the author puts it, most of the results watched dying "died at the second question rather than the first." Statistical significance and tradability are different questions, and the first is far easier to pass.
What did the pre-registered tests show?
The write-up's discipline is its most notable feature. Before running anything, the author wrote down ninety-six tests in three families, Bonferroni-corrected for their number. Zero passed. A frozen holdout sample was never opened, because nothing qualified to be tested against it.
The author argues this is what separates a result from a fishing expedition. Ninety-six tests with no passes is a result. Ninety-six tests searched afterwards for the three that worked is, in the author's words, "a paragraph about nothing." Declaring the tests before running them is what makes a zero mean something.
Whose reward is the reversion?
The write-up's own conclusion carries the mechanism: "nothing here is a trading signal for a taker; it is a liquidity provider's reward." The reversion is the market paying somebody for supplying liquidity. A trader who crosses the spread to enter a position is the one paying that reward, not collecting it.
The practical order of operations follows from this. Measure the effect. Then measure what it costs to collect it. Only then argue about entry rules. The author notes that most research, including plenty of their own, runs those steps in the opposite order — and spends months perfecting entry rules for effects that never cleared the cost hurdle.
The figures come from a research run over the stated instruments, windows and cost model. Historical simulations are not live results, and the author states that nothing in the write-up is investment advice. For quants and retail testers alike, the takeaway is structural: in markets this efficient at the one-minute horizon, the measured edge belongs to whoever stands still and quotes, not to whoever pays the spread to move.
Source: tradingview.com
More from Daniel Okafor
Show full bio
Correspondent covering money, funding and market shifts around entrepreneurship, startups and venture funding at Business Bearings.
2,400 articles