PHAITRADINGResearch
No market opinions here: only measurements, and nearly all of them say no. Every document has the same shape — the question, the method, the numbers without commentary, the verdict, and the measurement errors found along the way, which are often worth more than the result.
We publish what did not work — above all that — because the biggest cost in research is not trying new things: it is retrying things already dead. What stays private are the documents describing an engine in production.
Le inversioni non si addensano sui livelli tondi, nemmeno a risoluzione tick
The points where price turns do not cluster on round numbers, not even at one tick every four tenths of a second. But getting there meant throwing away two of my own measurements: one said the opposite with p = 1.0000, the other claimed a power the test did not have.
Falsified13 minAll'apertura della settimana lo spread si traveste da fuga verso i rifugi
A “risk-off” weekend gap at t +9.19: free money, therefore false. At the open the spread is 10.4 times normal, and the arithmetic can be done by hand.
Falsified11 minSei prove superate, e nessuna era quella giusta
Median, sign test, trimming, leave-one-out: three of six made it stronger. Then it met eleven years it had never seen and died in one stroke.
Falsified9 minLe candele non dicono niente, e il venerdì non esisteva prima del 2010
Candlesticks, round numbers and seasonality. The first two die as predicted; the third looked alive for half a day, then met eleven years it had never seen.
Falsified10 minL'ora del giorno era volatilità travestita
Six layers of analysis, three claims retracted while checking them, and in the end the hour of the day dissolves: it was never a cause, only a name for a state of the market.
Falsified7 minDove il momentum sembra funzionare, stare fermi funziona meglio
The one anomaly in this family with academic evidence. Where it seems to work, doing nothing works better: it was drift, not prediction.
Falsified7 minGli indicatori classici non aggiungono niente
Forty standard indicators, two million bars: fifty-seven columns collapse into nine dimensions, and on movement they lose to an elementary in-house average.
Falsified23 minIl cuore di Elliott è una media mobile, fatta peggio
Eleven hundred tests to force a theory to commit in advance. Elliott's fractal core is beaten by a 200-bar moving average.
Falsified7 minDow non aggiunge niente a non filtrare affatto
“Higher highs and higher lows” adds nothing over not filtering at all: p = 0.77. A moving average does the job, and better.
Falsified6 minIl voto dell'elezione non predice i soldi
Three ways to make the election cleaner, measured: where the cleaning succeeds, the noise seats disappear and risk goes down — money never does. And the literature had already written it.
Tool6 minIl backtest, spiegato riga per riga
The reading manual for our numbers: where they come from, what each one says — and the three things a backtest cannot say by construction.
La reversione ha un orologio? — protocollo pre-registrato
Zero out of twelve: reversion has no clock. Neither the session nor the hour separates the results from noise.
Falsified3 minI primi 10 minuti dopo l'evento, e la scia — protocollo pre-registrato
Neither the direction in the first ten minutes nor the wake in the half hour after: the event moves the market, but it does not say which way.
Falsified3 minLa notizia dà una direzione? — protocollo pre-registrato
Zero out of seventy-one types of news give a direction. The bias that seemed to be there came from dirty comparison windows: other events were inside them.
Falsified2 minIl regime separa i risultati? E dice da che parte stare?
The «6.1 times the error» was circularity: the bar it was measured on already contained the move that produced the result.
Falsified6 minIl regime: quanto si può ancora spremere
The refinements on regime fall one after another; what remains is a wrong piece of accounting, corrected.
Falsified5 minLa direzione vive nell'ora? E oltre le 24 ore? — protocollo pre-registrato
The hourly promotions were almost entirely a rollover artefact: the price at midnight is not a market move.
Falsified5 minLa direzione si predice in diretta? — protocollo pre-registrato
Pre-registered protocol: you declare what you will measure BEFORE looking. Direction, live, cannot be predicted — and the only one glimpsed sits ten to forty times below the spread.
Validated3 minIl Tester addebita uno spread che il conto non paga
The Tester charges a median spread of 0.20 pip. The real account pays far more. It is not a flaw in the strategy: it is a flaw in the measurement, and it touches everything measured on top of it.
Falsified3 minPiù celle per simbolo invece della sola migliore
Keeping the top two cells per symbol instead of only the best one. Measured: worse.
Falsified4 minL'impilamento senza stop: quanto spesso salta
Fourteen buys in three and a half hours, 93% closed in profit. Then the measure of ruin: fifty-one times past the account balance in sixteen years.
Falsified10 minQuanto deve durare un'operazione dell'Orologio
The gain matures over hours: cutting early pays the cost without collecting the drift. And the resolution test unmasks the stop that seemed to work.
Falsified3 minAggiungere posizioni dentro la finestra quando la cella è in rosso
Averaging down feels free until you look at the drawdown: up to 93.2% of the account. And the USDCHF signature shows why red, there, is not an opportunity.
Falsified3 minEleggere tutte le celle invece delle prime tre
If three cells work, will twenty-four work more? No: more cells means each one becomes smaller AND worse.