Intraday Trade Clustering: Resample Portfolio Days

Intraday trade clustering: Preserve shared sessions and portfolio equity instead of treating overlapping trades as independent samples.

The 7,137-trade discussion presented an intraday equities backtest and asked what to inspect first. Accessible comments proposed resampling whole trading days and comparing paper fills. The headline metrics and private rules were not independently reproduced. A large trade count is a reason to inspect dependence and accounting, not a substitute for that inspection.

The source discussion was reviewed through accessible rendered material. Accessible comments remain incomplete, and author results are unverified. The following experiment is original educational reasoning with hypothetical data. It does not reproduce the reported strategy or estimate its future performance.

Intraday trade clustering: intraday trade clustering: preserve shared dates

Start with a dated order and fill ledger. Identify every trade's session, quantity, costs and capital usage. Construct portfolio equity at a consistent observation frequency, including idle cash and open positions. Do not compound every position's percentage return as though the entire account entered each trade. Overlapping positions share capital and market exposure.

Next group activity by session. Record the number of trades and the largest simultaneous exposure. A session with many trades can still represent one shared directional episode. Keeping the session identifier provides a way to examine that concentration without pretending separate order identifiers create independent evidence.

Worked example: two positions, one account

Imagine an account begins with 1,000 cash. It allocates 100 to each of two positions, leaving 800 cash. Both positions gain 10%, so the account ends at 1,020 before costs: a 2% account return. Multiplying two 10% position returns would produce 21%, which answers a different and inappropriate capital-allocation question. These invented values make the accounting distinction visible.

Now imagine both positions gained because the same market event moved their prices. Shuffling the two position returns separately destroys their common date. Keeping the portfolio's session return preserves their realized joint effect under the supplied accounting. That does not prove the observed event is representative of future conditions.

Compare blocks as a sensitivity exercise

Build a small series of net daily equity returns from the ledger. Compare resampling individual days with resampling contiguous groups of days. Record the block lengths, seed, simulation count and path horizon before examining outputs. Keep unavailable histories explicit when a chosen block is unsupported.

Inspect drawdown and terminal-return differences across the assumptions. A longer block retains more observed local sequence, while offering fewer distinct starting episodes in a fixed sample. It is not automatically more accurate. Resampling cannot create an unseen halt, execution outage or structural change simply because the simulated path is long.

Check fills before interpreting the paths

Replay the same dates against any saved paper observations, retaining missing signals and unfilled orders. Separate delayed signals, spread assumptions and price concessions. A matching signal timestamp does not guarantee a matching position or fill. Keep this reconciliation separate from the resampling result so a favorable simulated quantile cannot conceal incorrect source accounting.

The useful outcome is a documented chain from fills to portfolio equity to conditional path diagnostics. It describes the supplied history under stated assumptions. It neither validates private strategy rules nor turns a simulated tail percentile into a real-world risk probability.

Reference and first useful step

Research selection depends on the number of attempted configurations. The Probability of Backtest Overfitting.

Explore a local sample in Advisory Research, then inspect your own dated equity ledger. Choose a synthetic sample before importing offline JSON. Read backtesting fundamentals and paper trading for background. Outputs remain advisory and do not execute trades.

Continue the research

Small Account Backtests: Test Capital Constraints and Backtesting Data: Choose the Resolution You Need develop adjacent checks. Use a small synthetic fixture first; these educational diagnostics do not validate a trading edge or change account settings.