Trade Backtester

Create trading strategies and run backtests. This package is still in development and is not meant to be used in production.

Contributors welcome: https://github.com/andyevers/trade-backtester

Development & Strategy Creation Rules

Speed is a top priority. Code can execute hundreds of thousands of times syncronously while testing hundreds of strategies, so every ms counts. Record performance checks after adding to code. generally, do 100,000 executions performance.test.ts

Function Usage

• Avoid Array functions that require reindexing (slice, shift, etc...) pretty much just allow pop()
• Avoid Math functions, just manually do calculations
• Avoid spread operator ({ ...variable })

Loops

• Never loop over all candles
• If searching for an item, consider using binary search (see PriceHistoryRepository.getIndexNearTime)
• Implement checks to skip unnecessary loops (see TriggerRepository.getTriggerLine)
• If filtering items, consider indexing them to avoid iterating over all items (see PositionRepository.getByIdLookup)

Misc

• Avoid creating huge indexes... I tried indexing candles by time and it slowed things down a lot. binary search ended up being a lot faster.
• Do not modify returned objects such as accounts, candles, positions, etc... this will alter the object stored in the repositories. This is because returning references to already created objects is much faster than cloning every object.

Notes

Additional Timeframes In Backtests

• Placing trades on additional timeframes will not be completely accurate. Unless the candle times are the same as the main candles, market orders will be placed on the next candle, or potentially expire if using orderDuration: DAY
• StopLosses will always fire before takeProfit if both are hit in the same candle, providing sub-timeframes for the symbol will not change this.
• Using multiple timeframes on the same symbol may cause accuracy issues. for example, if your main priceHistory is 'GM' 'day' and you provide AAPL 'month' and AAPL 'minute', the AAPL month and minute timeframes will fire in whatever order they were provided in the priceHistory array, activating triggers in unexpected orders.
• When using sub-timeframes, candles will be combined to a single candle when activating triggers or getting price data. for example, if your main timeframe is 'day' and you also have 'hour4', 6 hour4 candles will be combined to activate triggers and get current price data for that symbol.
• Triggers are activated while each new candle is being iterated, so when one trigger is activated (like a stopLoss or takeProfit), be aware that other candle data may not be updated yet.
• Each trigger is only tested once per symbol. if you have AAPL 'day' and AAPL 'minute', triggers on symbol AAPL will be tested only using the main timeframe if present, or the first combined candle for AAPL

Todo

• [x] TriggerService.processCandle is SUPER slow. fix this.
• [x] In Backtester.ts, add function to run a strategy (fires init and next).
• [ ] The term "Order" and "Position" are being used for the same thing (closeOrder returns Position etc...). Change to make terms uniform.
• [ ] Creating a candle copy in Timeline.candleGenerator takes about 10ms - 15ms for 70,000 iterations. This is only used to deal with current prices on alternate timeframes. find a faster way of doing this.
• [ ] inactive triggers are currently not in use. Delete deactivate triggers and related methods in TriggerRepository
• [ ] Remove event listeners from StrategyResultsAnalyzer and move to Backtester. Instead of using the addCandles event listener to watch each iteration, add a callback inside Backtester.runTest which should fire after strategy.next.
• [ ] In Timeline.ts, make reset() change things back to original state after settings start time. Keep in mind this may run hundreds or thousands of times syncronously while testing multiple strategies.
• [ ] Verify that all calculations are correct for StrategyResultsAnalyzer.ts. (especially sharpeRatio, calmarRatio, etc...)
• [ ] Take into consideration 252 trading days in a year when calculating returns
• [ ] Figure out what to do about positions opened but never closed in strategy analysis. This will mess with profit stats, expected returns, etc...