Out-of-sample testing: why trading results on unseen data matter
Out-of-sample testing scores a model only on data that was not used to build or tune it. It answers the question that matters: how does the model do on situations it has never seen? Results measured on training data (in-sample) are usually too optimistic.
In-sample versus out-of-sample
In-sample results describe how well a model fits the past it learned from. Out-of-sample results describe how well it handles new data. A large gap between the two is a classic sign of overfitting.
Questions to ask any tool
Was every reported number measured on data the model never saw? Were thresholds and settings fixed before testing? Was the list of markets chosen before seeing results? Is there a published method you can inspect? If a vendor cannot answer these, the headline accuracy is hard to trust.
How ApolloAlpha reports it
All headline figures are out-of-sample: 72.7% of weekly calls right, 82.6% at 70%+ confidence, AUC 0.799, Brier 0.183, across 163 markets. It is a historical test, not live trading, and accuracy is not profit.
Frequently asked questions
What is overfitting?
When a model learns noise in its training data and so performs worse on new data than its training results suggest.
Is out-of-sample the same as live trading?
No. It is still a historical test. Live results can differ because of costs, execution and changing markets.
Published 2026-11-07. See the track record and audit for the full method. Statistical model output, not financial advice. Past results do not guarantee future results.