Cluster Neutral Book Builder

A hand-picked list of "these pairs move together" grouping goes stale the moment correlations shift, and a book that is long-short by symbol count alone is not actually neutral if the two legs carry very different volatility or currency exposure. On each rebalance pass this EA builds a real Pearson correlation matrix from daily close-to-close returns across the whole candidate list, then clusters symbols using threshold union-find: any two symbols whose absolute correlation clears InpClusterThreshold are merged into the same cluster, and that merging propagates transitively (if A clusters with B and B clusters with C, all three end up in one cluster even if A and C alone do not clear the threshold). A symbol whose correlation to everything else falls short is left in a cluster of one and stays flat - there is nothing to pair it against. Inside a cluster of two or more, members are ranked by recent momentum. The weaker half is bought and the stronger half is sold, on the working assumption that a correlated cluster's members converge back toward each other over time. Each leg is first sized to an equal percentage-of-equity risk using its own ATR, then the short leg is scaled so the cluster's total short money-risk matches its total long money-risk as closely as lot stepping allows. Because every cluster is built to be individually neutral this way, the whole book - being a sum of neutral clusters - stays close to flat in aggregate without a separate whole-book balancing pass bolted on afterward. Existing positions are tracked purely by this EA's own magic number, ticket-collected on each pass; a symbol that drops out of a real cluster (now a singleton) has its position closed, and a symbol whose target side flips between passes is closed and reopened rather than left stale. Notes and limitations - Correlation and clustering are recomputed from scratch every rebalance pass, not incrementally - a symbol's cluster membership can genuinely change from one pass to the next as its recent returns change. - Clustering uses at most 100 correlation bars internally regardless of InpCorrelationBars, to keep the per-pass correlation matrix a bounded size; values above 100 are silently capped. - Cluster neutrality is money-risk neutral (ATR-weighted), not simple lot- count neutral - a cluster can hold more short lots than long lots (or vice versa) while still being risk-balanced, by design. - An odd-sized cluster is not perfectly balanced member-for-member; the scaling step balances risk, not headcount, between the two sides. - A symbol without enough loaded history for the configured InpCorrelationBars is excluded from clustering for that pass and stays in its own singleton (flat) rather than blocking the whole rebalance. - Existing same-direction positions are not resized between rebalances even if the target lot size has drifted; resizing only happens on the next full pass, or immediately on a direction flip. - No stop-loss is placed by design - risk control here is the ATR-based sizing and the risk-parity balancing, not a price stop.

Expert advisor · MT5 · v1.0 · Free

Tags: Risk & Lot Size · ATR · MT5 Expert Advisor · MQL5 Source Code

Expert advisors (EA) for MT4 & MT5 · PolyMaria