Walk Forward

Walk Forward is a JavaScript framework that lets you backtest your trading strategies (stocks, futures, forex, crypto currencies, …).

Features

• Uses re-usable algorithms (signal generation, position sizing, risk management etc.) that can be freely stacked together.
• Let's you plug in any data source you like – CSVs, web services, a broker's API etc. (only supports CSVs out of the box for now)
• Handles any instrument you wish to backtest: Futures, Stocks, Crypto Currencies, Forex etc.
• Produces standardized CSV output that can be read and displayed by any CSV-capable application (Google Spreadsheets, Microsoft Excel; a matching frontend is on the roadmap)
• Parameter optimization (with log scales)
• Written in a flexbile, heavily used and steadily improving language (yes, JavaScript), using current developments (ES6 modules, async/await)
• Used and tested (unit/integration tests) on the server side, but should also work in the frontend
• Open source (ISC license)
• Uses tulip's indicators

Example Code

Prerequisites

1. You need Node.js and NPM.
2. Install WalkForward: npm i -S walk-forward
3. WalkForward uses ES2015 modules (it's the future). In order to use them in current Node.js versions (below 10), we have to transform them to CommonJS modules. We use babel-register to do so: npm i -S babel-register

Your Code

Instruments (TransformableDataSeries)

The foundation of WalkForward are TransformableDataSeries. They're used for instruments, positions and accounts (cash/positions etc.).

Indicators

Algorithms

WalkForward simulates the strategy you provide based on the data you provide and outputs the corresponding statistics (basically gains and losses). The strategy is called whenever an instrument closes and is expected to return orders. Orders tell WalkForward what, when and how much to buy. A strategy can be split up into different parts (algorithms):

1. Trading signals: Should we buy now?
2. Position Sizing: If we buy, how much should we buy?
3. Risk management: Make sure we don't bet everything on one horse.

As those algorithms can by their nature be used in different strategies, WalkForward makes them re-usable and pluggable.

A basic algorithm looks like this:

import { SMA } from 'walk-forward-indicators';

/**
 * Buys 
 */
class SMAStrategy {

	/**
	 * onNewInstrument is called by WalkForward whenever a new instrument is discovered in the
	 * data you provide.
	 */
	onNewInstrument(instrument) {
		// instrument.addTransformer returns a Symbol() that can be used to access the transformer's
		// values when the backtest is running (see onClose method)
		this.fastSMA = instrument.addTransformer(new SMA(5));
		this.slowSMA = instrument.addTransformer(new SMA(10));
	}

	/**
	 * onClose is called by WalkForward whenever data (open/close etc.) was received for an 
	 * instrument. 
	 * It is expected to return an array of orders; every order is an object with two properties:
	 * - size (number) 
	 * - instrument (instrument)
	 * onClose is called with the following arguments: 
	 * - orders (current orders)
	 * - instrument (the instrument which received the data and the close event was fired on)
	 * - instruments (an array of all instruments)
	 * - accounts (the current account)
	 */
	onClose(orders, instrument) {
		if (
			// Current fastSMA is above current slowSMA
			instrument.head().get(this.fastSMA) > instrument.head().get(this.slowSMA) &&
			// Previous fastSMA was below or equal to previous slowSMA; this makes sure that 
			// we only buy on cross overs.
			instrument.back(1).get(this.fastSMA) <= instrument.back(1).get(this.slowSMA)
		) {
			// [...orders] clones the orders array (as you should not modify parameters passed
			// to your functions). It is then extended with the current order.
			// We use a default size of 1 which will later be modified by the position sizing
			// algorithm
			return [...orders, { instrument: instrument, size: 1 }]
		}
		// Default: Just return the existing orders. If we returned an empty array, all existing
		// orders would be canceled.
		return orders;
	}
}

Backtest

In order to run a backtest,

1. instantiate a new Backtest()
2. provide a data source (e.g. some CSV files)
3. write an algorithm that creates the buy signals and another algorithm that

Documentation

Logging

To enable logs, set the log level and components through your environment:

export DEBUG=WalkForward:*
export "DEBUG_LEVELS=debug, info, warn, error"

Algos and stacks

Algos

Backtest

backtest.run()

Transformers

Instruments, results

Instruments

Roadmap

• [x] Parameter optimization
• [ ] Configure spread, commission and slippage depending on an order
• [x] Output relevant display data as JSON (for formatted output in frontend)
• [x] Interactive web frontend
• [ ] Instrument configuration (type of instrument, whole sizes, margins)
• [ ] Margin handling
• [ ] More optimization techniques (in addition to current log, e.g. linear, other log bases, integers only instead of floats)
• [x] Relevant ouptut in form of tables (orders, positions in addition to charts)
• [ ] Better charts for parameter optimization (2D/3D visualization)
• [ ] Currency handling
• [ ] Walk forward optimization
• [ ] More data sources out of the box - Better docs, always better docs

Inspiration

koa's Middleware, bt's reusable algos, glulp 4's chainable serial/parallel functions.