What this evidence does not establish.
The literature supports part of the logic. The rest is what Graham Says is testing.
Price discovery is not closing-level predictability.
The cited literature studies leads measured in seconds or minutes. Graham Says asks whether a pre-open futures signal contains information about a cash-index close many hours later. Whether that signal helps at a daily horizon is a separate question tested by this project.
1 · A small live sample is not a track record
Early prospective errors are observations, not stable estimates of forecasting skill.
2 · The literature is mostly old and intraday
The core studies use samples from the 1980s through 2000. Market structure has changed substantially since then.
3 · Futures do not lead cash on every occasion
The relation is an average tendency rather than a law, and some studies report feedback in the other direction.
4 · Composite and Nasdaq-100 are different indices
Any divergence between them is a source of forecast error.
5 · The futures measurement has instrument risk
Contract rolls, the futures basis and varying overnight liquidity can all matter. The pipeline explicitly guards against contract mismatch.
6 · A frozen model cannot adapt
That is deliberate: it creates a cleaner out-of-sample test, but it means the model cannot quietly adjust if the relationship changes.