Low-latency signal replication with 90% synchronisation accuracy
How we built a real-time trading signal replication engine that reduced execution delay by 70% and near-eliminated trade mismatch errors across multiple accounts.
The situation
A proprietary trading firm running a signal account wanted to replicate trades automatically across multiple follower accounts. The existing approach — traders manually watching the signal account and copying positions — introduced delays of anywhere from 5 to 90 seconds depending on how quickly a human noticed a new trade. At the execution prices involved, a 30-second delay on a fast-moving instrument could mean several hundred points of slippage.
Mismatches between the signal account's position and follower accounts were common. A partial fill on the signal account would sometimes be replicated as a full fill, or a position would be closed on one follower but not on others due to a missed WebSocket message. Reconciliation was a manual process run at day-end.
What we built
Signal detection engine
A persistent WebSocket connection to the broker's trade event stream monitors the signal account for new trades, modifications, and closures in real time. Each event is normalised into a canonical signal format and pushed to a Redis queue within milliseconds of detection.
Multi-account replication layer
A worker pool processes signals from the Redis queue and fires execution requests to each follower account's broker API in parallel. Worker count scales with follower account count — adding a new follower account is a database record, not a code change. Each execution attempt is logged with timestamp, order ID, and fill price.
Position reconciliation engine
A background reconciliation job runs every 60 seconds, comparing the signal account's open positions against each follower account. Discrepancies (missing positions, incorrect sizes) surface as alerts in the monitoring dashboard and can optionally trigger automatic correction orders.
Monitoring and audit dashboard
A real-time dashboard showing signal account activity, each follower's replication status, execution latency per trade, fill prices vs. signal price (slippage tracking), and a reconciliation summary. Historical trade data queryable for post-analysis.
Technical decisions worth explaining
Redis queue for durability under burst load
During high-volatility market events, signal volume can spike dramatically (multiple trades in seconds). Using Redis as a durable queue between detection and execution means no signals are dropped even if the execution workers are temporarily saturated. Workers drain the queue in order as capacity allows.
Idempotent execution
Each signal is tagged with a unique ID derived from the source trade ID and event type. The execution layer checks for a matching record before placing an order, ensuring network retries or duplicate WebSocket messages can never cause a duplicate trade.
Results
- ~70% reduction in signal execution delay (human copy-trading → automated replication)
- ~90% improvement in synchronisation accuracy across follower accounts
- ~60% reduction in trade mismatch errors
- Day-end reconciliation replaced by continuous automated reconciliation with real-time alerts
- Follower account onboarding reduced from a development task to a database record