Alpha is return above what a chosen benchmark (and risk model) would explain — useful as the language for “skill” or edge in portfolio and trading discussions, separate from market beta.
Whilst alpha is sold as pure skill, in practice it collapses under wrong benchmarks, ignored costs, and overfitting. For example, a backtest beats the index until fees, slippage, and regime change erase the edge. We often recommend reporting net of costs, fixing the benchmark up front, and treating live implementation shortfall as part of the alpha story — not a footnote.
Alpha is a model-relative number: return minus what risk factors predict. Without a declared benchmark, risk model, and cost stack, “we make alpha” is marketing. In execution systems, the fight is often preserving theoretical edge through latency, fills, and risk limits.