Back to Home

Methodology

True Alpha is a scoreboard for finance creator predictions — built to answer: who is actually beating the market, and is it statistically meaningful?

1) Data Collection

We pull predictions from creator videos using transcripts. An AI model extracts: ticker symbols, direction (bullish/bearish), confidence, and a target horizon when available.

Important: transcripts and AI extraction can be imperfect. We treat this as a best-effort data engine, not a brokerage ledger.

2) Entry Price (Your Rule)

Entry price is the close price on the publish date. This is the simplest, most repeatable assumption across creators and time zones.

3) Evaluation Horizon

Each prediction is evaluated using a 365-day window by default (publish-date close → same date next year).

The leaderboard also breaks out shorter windows (90/180 days) where available.

4) Alpha (What "True Alpha" Means)

Returns are meaningless without a benchmark. We compute Alpha as: Stock Return − S&P 500 Return over the same horizon.

Alpha = (Stock % Return) − (S&P 500 % Return)

5) Accuracy

Accuracy is directional: if a creator is bullish and the asset is up over the evaluation window, that's a "win." If they're bearish and it's down, that's also a "win."

6) Statistical Significance

We run a "skill vs luck" check on every creator. If their performance is strong enough — across enough predictions — that it's unlikely to be random, we mark them as Statistically Significant.

We use a one-sample t-test against a null hypothesis of zero alpha. Creators with p < 0.05 earn the "Sig" badge on the leaderboard.