Professional research is about uncertainty, execution, samples and reproducibility. These four original training notes are designed for self-assessment, not for following trade calls.
01 / MARKET MICROSTRUCTUREA spread is a state variable, not a fixed fee.
A nominal 1.2-pip spread cannot be assumed to persist through releases or thin liquidity. Model the bid/ask side, expected execution depth, partial fills and the difference between a stop trigger and an executable price.
Research exercise: record the spread at observation time and after a volatility shock; recompute net expectancy under a doubled spread and one additional adverse fill increment. Mark the data source and its latency.
02 / PORTFOLIO INTERDEPENDENCEFive small positions can be one large idea.
Correlated exposure can compound at regime transitions. Two positions with 1% independently budgeted losses are not diversified just because their tickers differ; overlapping factors and tail dependence matter.
Research exercise: identify shared currency, rate, commodity and market-beta factors. Stress simultaneous stop-outs plus a gap loss, then compare the total with your self-chosen ceiling.
03 / WALK-FORWARD VALIDATIONOut-of-sample does not mean future-proof.
Repeated parameter searches contaminate holdout sets. Define research and test periods in advance, track strategy revisions and separate simulation, forward demo and broker-reconciled outcomes. Include commissions and sample uncertainty.
Research exercise: document the number of variations tested; show chronological train/test splits and investigate sensitivity to doubled costs and poorer fills.
04 / REGIME-CONDITIONAL EXPECTANCYAn average can hide incompatible subgroups.
A +0.15R aggregate sample may be driven by a single market regime. Segment by volatility, session, signal family and holding period. Report sample sizes and confidence limits rather than implying that a small positive sample proves a trading edge.
Research exercise: calculate expected R separately in calm and high-volatility cohorts, with the same fee model. Flag fewer than 30 observations as a small sample, not as verified performance.