Does your indicator know anything? The information coefficient, explained with real numbers
An indicator can look convincing on a chart and still say nothing about the next candle. The information coefficient measures how well it ranks future returns. What it is, how to read it, and what ten popular indicators scored.
An indicator can look convincing on a chart and still say nothing about what price does next. The line bends at the right places because it is computed from the same prices you are looking at. The question worth asking is different: when the indicator is high, does price tend to do better afterwards than when it is low?
The information coefficient
The information coefficient (IC) answers exactly that. For every candle we take the indicator value and the return over the next h candles, rank both, and compute the correlation between the ranks (Spearman correlation). Ranking makes it robust to outliers: one crash day cannot dominate the result.
- IC near 0 — the indicator's value tells you nothing about the next move.
- Positive IC — higher values are followed by better returns.
- Negative IC — higher values are followed by worse returns (still useful, the other way round).
In markets, useful ICs are small. An IC of 0.05 on daily data can matter; an IC of 0.5 almost certainly means a bug, usually the indicator peeking at the future.
Is it more than noise?
A small IC on a small sample is meaningless, so we also compute a t-statistic. Returns over overlapping windows are not independent — a 20-day return today and tomorrow share 19 days — so we count the effective sample as n / h, which is stricter. Evident also splits the data into two halves: an indicator that works in one half and reverses in the other is not stable, and gets a weaker verdict.
What ten popular indicators scored
We measured RSI, distance from the 50-day EMA, an EMA 20/50 cross, the MACD histogram, Bollinger %B, ATR as a percentage of price, 20-day momentum, a volume z-score, a 20-day Donchian breakout and a 50-day price z-score on BTC, ETH, SOL and SPY daily candles. In the 27 September 2026 measurement:
| Indicator, market | IC (next candle) | t |
|---|---|---|
| Donchian 20-day breakout, SOL | 0.064 | 2.40 |
| Donchian 20-day breakout, BTC | 0.058 | 2.16 |
| ATR % of price, SPY | 0.071 | 2.18 |
| Volume z-score, SPY | 0.069 | 2.11 |
| 50-day price z-score, SPY | −0.081 | −2.45 |
Those are the 5 of 40 combinations that passed our evidence bar; 3 more were weak and 32 showed nothing. And here is the important caveat: with 40 tests at the 5% threshold, about 2 "findings" are expected from chance alone. Five is more than two, so something may be there — the breakout result appearing on two different crypto markets is more convincing than any single line — but each of these is a hypothesis to test in a strategy with costs, not a proven edge.
Notice also what did not make the list: RSI, MACD and moving-average crosses, the indicators most strategies are built on, carried no clear information about next-day returns on these markets.
Measure your own
In Evident's indicator lab every indicator — one you describe to the AI, one you paste from Pine Script, or one you write yourself — is measured this way on the market and timeframe you choose, with IC at several horizons, quintile returns and the two-halves stability check. The live indicator study is recomputed every day.
Research, not investment advice.
Evident does not give investment advice. Past performance does not indicate future returns. Leveraged and automated trading carries the risk of losing capital.