413
Markets tracked
We do not forecast price. We measure when a market behaves differently — and discard everything that fails to beat chance.
Every asset faces the same examination: all possible combinations of entry and exit hour, in both directions, with execution cost deducted from the first calculation.
1,104 windows per asset across 200 days of hourly data. Each is scored on net PnL, win rate, drawdown and statistical significance.
The strongest twenty assets in the universe, replaced only when a candidate beats the weakest by a margin. No turnover from noise.
The book runs against real prices, position by position, with state persisted daily and returns compounded on the running balance.
Relative value, correlation, hedging and statistical anomaly — each with its own bounded search space.
Testing a thousand hypotheses guarantees winners by chance. Every sweep compares its best t-statistic against what pure noise would produce at that number of trials.
Each window is scored separately across both halves of the period. One that only works where it was found is never marked robust.
We subtract what plain exposure would have returned over the same hours. A rising asset makes any long strategy look good — that is not a signal.
Gross results are never presented as attainable. Execution cost enters the score itself, not a footnote beneath it.
Sweeps, portfolios, live execution and a daily record of results.