Signal Backtest

Proof, not promises. This page replays PokéAlpha's Fair Value discounts and Opportunity Scores against history to measure whether high-scoring assets actually went on to outperform — by hit-rate, realized return, and the equity curve of acting on the top signals.

Do the signals predict outperformance?

no clear edge

Over this window the signals did not reliably predict outperformance — treat the scores as one input, not a guarantee.

Measured over 40 weekly as-of dates across 360 tracked assets (28,800 point-in-time signals), from Jan 1 to Oct 3.

Hit-rate & realized returns

How often a high-conviction signal (top quintile by Opportunity Score on each date) beat the equal-weight market, and the average benchmark-relative return it earned.

30d hit-rate (high)

50%

beat the market (2,880 signals)

30d avg excess (high)

+2.4%

vs equal-weight market

90d hit-rate (high)

49%

beat the market (2,880 signals)

90d avg excess (high)

+4.2%

vs equal-weight market

High-conviction vs low-conviction

If the score is real, the top cohort should out-earn the bottom cohort over the same window. The long/short spread is that gap in benchmark-relative return.

Horizon	Cohort	Signals	Avg return	vs market	Hit-rate	Long/short
30-day	High conviction	2880	+3.6%	+2.4%	50%	+4.3%
30-day	Low conviction	2880	-0.7%	-1.9%	41%	+4.3%
90-day	High conviction	2880	+6.2%	+4.2%	49%	+7.3%
90-day	Low conviction	2880	-1.1%	-3.1%	40%	+7.3%

Equity curve — acting on the top signals

A portfolio that rebalances monthly into the 15 highest-scoring assets (equal-weighted, net of turnover cost) versus simply holding the whole market.

Act on top signals 157Buy the whole market 113Both indexed to 100 at the start.

Strategy return

+56.6%

Top-signal portfolio, total

Market return

+13.3%

Equal-weight benchmark, total

Max drawdown

-3.0%

Worst peak-to-trough

Sharpe ratio

3.52

Annualized, monthly periods

Information coefficient

The Spearman rank correlation between a signal's score and its realized forward return. 0 means no predictive power; a small positive value (0.03–0.10) is a genuine, usable edge in real-world quant terms.

30-day IC

0.088

across 14,400 signals

90-day IC

0.107

across 14,400 signals

Methodology & honesty

A backtest is only trustworthy if it can't cheat. Here is exactly how these numbers are produced — point-in-time, with no look-ahead, measured against the market, not in a vacuum.

Point-in-time: at each weekly as-of date the Fair Value and Opportunity Score are recomputed using ONLY the price/comp history available up to that date — no future data leaks into a signal.
No look-ahead: realized returns are read forward from the same continuous price series the signal was scored on, 30 and 90 calendar days later.
Benchmark-relative: every outcome is measured against the equal-weight return of the whole tracked universe over the identical window, so a rising tide doesn't masquerade as alpha.
Hit-rate counts how often a high-conviction signal beat that benchmark; the Information Coefficient is the Spearman rank correlation between score and forward return.
The equity curve rebalances monthly into the top-scoring assets, equal-weighted and net of a turnover cost, versus holding the whole market.
A genuine edge is modest and noisy: a hit-rate in the high-50s to low-60s is meaningful, not a crystal ball. This backtest runs over PokéAlpha's deterministic market model, not live trading.