autotrader/backtesting.go
2023-05-18 20:17:29 -05:00

645 lines
17 KiB
Go

package autotrader
import (
"errors"
"fmt"
"log"
"os"
"strconv"
"strings"
"text/tabwriter"
"time"
"github.com/go-echarts/go-echarts/v2/charts"
"github.com/go-echarts/go-echarts/v2/components"
"github.com/go-echarts/go-echarts/v2/opts"
"golang.org/x/exp/rand"
"golang.org/x/exp/slices"
)
var (
ErrEOF = errors.New("end of the input data")
ErrNoData = errors.New("no data")
ErrPositionClosed = errors.New("position closed")
)
var _ Broker = (*TestBroker)(nil) // Compile-time interface check.
func Backtest(trader *Trader) {
switch broker := trader.Broker.(type) {
case *TestBroker:
rand.Seed(uint64(time.Now().UnixNano()))
trader.Init() // Initialize the trader and strategy.
start := time.Now()
for !trader.EOF {
trader.Tick() // Allow the trader to process the current candlesticks.
broker.Advance() // Give the trader access to the next candlestick.
}
trader.closeOrdersAndPositions() // Close any outstanding trades now.
log.Printf("Backtest completed on %d candles. Opening report...\n", trader.Stats().Dated.Len())
stats := trader.Stats()
// Divide net profit by maximum drawdown to get the profit factor.
var maxDrawdown float64
stats.Dated.Series("Drawdown").ForEach(func(i int, val any) {
f := val.(float64)
if f > maxDrawdown {
maxDrawdown = f
}
})
profit := stats.Dated.Float("Profit", -1)
profitFactor := stats.Dated.Float("Profit", -1) / maxDrawdown
maxDrawdownPct := 100 * maxDrawdown / stats.Dated.Float("Equity", 0)
// Print a summary of the statistics to the console.
{
w := tabwriter.NewWriter(os.Stdout, 0, 0, 1, ' ', 0)
fmt.Fprintln(w)
fmt.Fprintf(w, "Net Profit:\t$%.2f (%.2f%%)\t\n", profit, 100*profit/stats.Dated.Float("Equity", 0))
fmt.Fprintf(w, "Profit Factor:\t%.2f\t\n", profitFactor)
fmt.Fprintf(w, "Max Drawdown:\t$%.2f (%.2f%%)\t\n", maxDrawdown, maxDrawdownPct)
fmt.Fprintln(w)
w.Flush()
}
// Pick a datetime layout based on the frequency.
dateLayout := time.DateTime
if strings.Contains(trader.Frequency, "S") { // Seconds
dateLayout = "15:04:05"
} else if strings.Contains(trader.Frequency, "H") { // Hours
dateLayout = "2006-01-02 15:04"
} else if strings.Contains(trader.Frequency, "D") || trader.Frequency == "W" { // Days or Weeks
dateLayout = time.DateOnly
} else if trader.Frequency == "M" { // Months
dateLayout = "2006-01"
} else if strings.Contains(trader.Frequency, "M") { // Minutes
dateLayout = "01-02 15:04"
}
page := components.NewPage()
// Create a new line balChart based on account equity and add it to the page.
balChart := charts.NewLine()
balChart.SetGlobalOptions(
charts.WithTitleOpts(opts.Title{
Title: "Balance",
Subtitle: fmt.Sprintf("%s %s %T %s (took %.2f seconds)", trader.Symbol, trader.Frequency, trader.Strategy, time.Now().Format(time.DateTime), time.Since(start).Seconds()),
}),
charts.WithTooltipOpts(opts.Tooltip{
Show: true,
Trigger: "axis",
TriggerOn: "mousemove|click",
}),
charts.WithYAxisOpts(opts.YAxis{
AxisLabel: &opts.AxisLabel{
Show: true,
Formatter: "${value}",
},
}),
charts.WithLegendOpts(opts.Legend{
Show: true,
Selected: map[string]bool{"Equity": false, "Profit": true},
}))
balChart.SetXAxis(seriesStringArray(stats.Dated.Dates(), dateLayout)).
AddSeries("Equity", lineDataFromSeries(stats.Dated.Series("Equity"))).
SetSeriesOptions(
charts.WithMarkPointNameTypeItemOpts(
opts.MarkPointNameTypeItem{Name: "Peak", Type: "max", ItemStyle: &opts.ItemStyle{
Color: balChart.Colors[1],
}},
opts.MarkPointNameTypeItem{Name: "Drawdown", Type: "min", ItemStyle: &opts.ItemStyle{
Color: balChart.Colors[3],
}},
),
)
balChart.AddSeries("Profit", lineDataFromSeries(stats.Dated.Series("Profit")))
// Create a new kline chart based on the candlesticks and add it to the page.
kline := newKline(trader.data, stats.Dated.Series("Trades"), dateLayout)
// Sort Returns by value.
// Plot returns as a bar chart.
returnsSeries := stats.Dated.Series("Returns")
returns := make([]float64, 0, returnsSeries.Len())
// returns := stats.Dated.Series("Returns").Values()
// Remove nil values.
for i := 0; i < returnsSeries.Len(); i++ {
r := returnsSeries.Value(i)
if r != nil {
returns = append(returns, r.(float64))
}
}
// Sort the returns.
slices.Sort(returns)
// Create the X axis labels for the returns chart based on length of the returns slice.
returnsLabels := make([]int, len(returns))
for i := range returns {
returnsLabels[i] = i + 1
}
returnsBars := make([]opts.BarData, len(returns))
for i, r := range returns {
returnsBars[i] = opts.BarData{Value: r}
}
var avg float64
for _, r := range returns {
avg += r
}
avg /= float64(len(returns))
returnsAverage := make([]opts.LineData, len(returns))
for i := range returnsAverage {
returnsAverage[i] = opts.LineData{Value: avg}
}
returnsChart := charts.NewBar()
returnsChart.SetGlobalOptions(
charts.WithTitleOpts(opts.Title{
Title: "Returns",
Subtitle: fmt.Sprintf("Average: $%.2f", avg),
}),
charts.WithYAxisOpts(opts.YAxis{
AxisLabel: &opts.AxisLabel{
Show: true,
Formatter: "${value}",
},
}))
returnsChart.SetXAxis(returnsLabels).
AddSeries("Returns", returnsBars)
returnsChartAvg := charts.NewLine()
returnsChartAvg.SetGlobalOptions(charts.WithTitleOpts(opts.Title{
Title: "Average Returns",
}))
returnsChartAvg.SetXAxis(returnsLabels).
AddSeries("Average", returnsAverage, func(s *charts.SingleSeries) {
s.LineStyle = &opts.LineStyle{
Width: 2,
}
})
returnsChart.Overlap(returnsChartAvg)
// TODO: Use Radar to display performance metrics.
// Add all the charts in the desired order.
page.PageTitle = "Backtest Report"
page.AddCharts(balChart, kline, returnsChart)
// Draw the page to a file.
f, err := os.Create("backtest.html")
if err != nil {
panic(err)
}
page.Render(f)
f.Close()
// Open the chart in the default browser.
if err := Open("backtest.html"); err != nil {
panic(err)
}
default:
log.Fatalf("Backtesting is only supported with a TestBroker. Got %T", broker)
}
}
func newKline(dohlcv Frame, trades Series, dateLayout string) *charts.Kline {
kline := charts.NewKLine()
x := make([]string, dohlcv.Len())
y := make([]opts.KlineData, dohlcv.Len())
for i := 0; i < dohlcv.Len(); i++ {
x[i] = dohlcv.Date(i).Format(dateLayout)
y[i] = opts.KlineData{Value: [4]float64{
dohlcv.Open(i),
dohlcv.Close(i),
dohlcv.Low(i),
dohlcv.High(i),
}}
}
marks := make([]opts.MarkPointNameCoordItem, 0)
for i := 0; i < trades.Len(); i++ {
if slice := trades.Value(i); slice != nil {
for _, trade := range slice.([]TradeStat) {
color := "green"
rotation := float32(0)
if trade.Units < 0 {
color = "red"
rotation = 180
}
if trade.Exit {
color = "black"
}
marks = append(marks, opts.MarkPointNameCoordItem{
Name: "Trade",
Value: fmt.Sprintf("%v units", trade.Units),
Coordinate: []interface{}{x[i], y[i].Value.([4]float64)[1]},
Label: &opts.Label{
Show: true,
Position: "inside",
},
ItemStyle: &opts.ItemStyle{
Color: color,
},
Symbol: "arrow",
SymbolRotate: rotation,
SymbolSize: 25,
})
}
}
}
kline.SetGlobalOptions(
charts.WithTitleOpts(opts.Title{
Title: "Trades",
Subtitle: fmt.Sprintf("Showing %d candles", dohlcv.Len()),
}),
charts.WithXAxisOpts(opts.XAxis{
SplitNumber: 20,
}),
charts.WithYAxisOpts(opts.YAxis{
Scale: true,
}),
charts.WithTooltipOpts(opts.Tooltip{ // Enable seeing details on hover.
Show: true,
Trigger: "axis",
TriggerOn: "mousemove|click",
}),
charts.WithDataZoomOpts(opts.DataZoom{ // Support zooming with scroll wheel.
Type: "inside",
Start: 0,
End: 100,
XAxisIndex: []int{0},
}),
charts.WithDataZoomOpts(opts.DataZoom{ // Support zooming with bottom slider.
Type: "slider",
Start: 0,
End: 100,
XAxisIndex: []int{0},
}),
)
kline.SetXAxis(x).AddSeries("Price Action", y, charts.WithMarkPointNameCoordItemOpts(marks...))
return kline
}
func lineDataFromSeries(s Series) []opts.LineData {
if s == nil || s.Len() == 0 {
return []opts.LineData{}
}
data := make([]opts.LineData, s.Len())
for i := 0; i < s.Len(); i++ {
data[i] = opts.LineData{Value: Round(s.Value(i).(float64), 2)}
}
return data
}
func seriesStringArray(s Series, dateLayout string) []string {
if s == nil || s.Len() == 0 {
return []string{}
}
data := make([]string, s.Len())
for i := 0; i < s.Len(); i++ {
switch val := s.Value(i).(type) {
case time.Time:
data[i] = val.Format(dateLayout)
case string:
data[i] = fmt.Sprintf("%q", val)
default:
data[i] = fmt.Sprintf("%v", val)
}
}
return data
}
// TestBroker is a broker that can be used for testing. It implements the Broker interface and fulfills orders
//
// Signals:
// - Tick(nil) - Called when the broker ticks.
// - OrderPlaced(Order) - Called when an order is placed.
// - OrderFilled(Order) - Called when an order is filled.
// - OrderCanceled(Order) - Called when an order is canceled.
// - PositionClosed(Position) - Called when a position is closed.
// - PositionModified(Position) - Called when a position changes.
type TestBroker struct {
SignalManager
DataBroker Broker
Data *DataFrame
Cash float64
Leverage float64
Spread float64 // Number of pips to add to the price when buying and subtract when selling. (Forex)
Slippage float64 // A percentage of the price to add when buying and subtract when selling.
candleCount int // The number of candles anyone outside this broker has seen. Also equal to the number of times Candles has been called.
orders []Order
positions []Position
}
func NewTestBroker(dataBroker Broker, data *DataFrame, cash, leverage, spread float64, startCandles int) *TestBroker {
return &TestBroker{
DataBroker: dataBroker,
Data: data,
Cash: cash,
Leverage: Max(leverage, 1),
Spread: spread,
Slippage: 0.005, // Price +/- 0.5%
candleCount: Max(startCandles, 1),
}
}
// CandleIndex returns the index of the current candle.
func (b *TestBroker) CandleIndex() int {
return Max(b.candleCount-1, 0)
}
// Advance advances the test broker to the next candle in the input data. This should be done at the end of the
// strategy loop.
func (b *TestBroker) Advance() {
if b.candleCount < b.Data.Len() {
b.candleCount++
}
}
// Bid returns the price a seller pays for the current candle.
func (b *TestBroker) Bid(_ string) float64 {
return b.Data.Close(b.CandleIndex())
}
// Ask returns the price a buyer pays for the current candle.
func (b *TestBroker) Ask(_ string) float64 {
return b.Data.Close(b.CandleIndex()) + b.Spread
}
// Candles returns the last count candles for the given symbol and frequency. If count is greater than the number of candles, then a dataframe with zero rows is returned.
//
// If the TestBroker has a data broker set, then it will use that to get candles. Otherwise, it will return the candles from the data that was set. The first call to Candles will fetch candles from the data broker if it is set, so it is recommended to set the data broker before the first call to Candles and to call Candles the first time with the number of candles you want to fetch.
func (b *TestBroker) Candles(symbol string, frequency string, count int) (*DataFrame, error) {
start := Max(Max(b.candleCount, 1)-count, 0)
adjCount := b.candleCount - start
if b.Data != nil && b.candleCount >= b.Data.Len() { // We have data and we are at the end of it.
return b.Data.Copy(-count, -1).(*DataFrame), ErrEOF // Return the last count candles.
} else if b.DataBroker != nil && b.Data == nil { // We have a data broker but no data.
candles, err := b.DataBroker.Candles(symbol, frequency, count)
if err != nil {
return nil, err
}
b.Data = candles
} else if b.Data == nil { // Both b.DataBroker and b.Data are nil.
return nil, ErrNoData
}
return b.Data.Copy(start, adjCount).(*DataFrame), nil
}
func (b *TestBroker) MarketOrder(symbol string, units float64, stopLoss, takeProfit float64) (Order, error) {
if b.Data == nil { // The DataBroker could have data but nobody has fetched it, yet.
if b.DataBroker == nil {
return nil, ErrNoData
}
_, err := b.Candles("", "", 1) // Fetch data from the DataBroker.
if err != nil {
return nil, err
}
}
var price float64
if units < 0 {
price = b.Bid("")
} else {
price = b.Ask("")
}
slippage := rand.Float64() * b.Slippage * price
price += slippage - slippage/2 // Get a slippage as +/- 50% of the slippage.
order := &TestOrder{
id: strconv.Itoa(rand.Int()),
leverage: b.Leverage,
position: nil,
price: price,
symbol: symbol,
stopLoss: stopLoss,
takeProfit: takeProfit,
time: time.Now(),
orderType: MarketOrder,
units: units,
}
// Instantly fulfill the order.
order.position = &TestPosition{
broker: b,
closed: false,
entryPrice: price,
id: strconv.Itoa(rand.Int()),
leverage: b.Leverage,
symbol: symbol,
stopLoss: stopLoss,
takeProfit: takeProfit,
time: time.Now(),
units: units,
}
b.Cash -= order.position.EntryValue()
b.orders = append(b.orders, order)
b.positions = append(b.positions, order.position)
b.SignalEmit("OrderPlaced", order)
return order, nil
}
func (b *TestBroker) NAV() float64 {
nav := b.Cash
// Add the value of open positions to our NAV.
for _, position := range b.positions {
if !position.Closed() {
nav += position.Value()
}
}
return nav
}
func (b *TestBroker) PL() float64 {
var pl float64
for _, position := range b.positions {
pl += position.PL()
}
return pl
}
func (b *TestBroker) OpenOrders() []Order {
orders := make([]Order, 0, len(b.orders))
for _, order := range b.orders {
if !order.Fulfilled() {
orders = append(orders, order)
}
}
return orders
}
func (b *TestBroker) OpenPositions() []Position {
positions := make([]Position, 0, len(b.positions))
for _, position := range b.positions {
if !position.Closed() {
positions = append(positions, position)
}
}
return positions
}
func (b *TestBroker) Orders() []Order {
return b.orders
}
func (b *TestBroker) Positions() []Position {
return b.positions
}
type TestPosition struct {
broker *TestBroker
closed bool
entryPrice float64
closePrice float64 // If zero, then position has not been closed.
id string
leverage float64
symbol string
stopLoss float64
takeProfit float64
time time.Time
units float64
}
func (p *TestPosition) Close() error {
if p.closed {
return ErrPositionClosed
}
p.closed = true
if p.units < 0 {
p.closePrice = p.broker.Ask("") // Ask because we are short so we have to buy.
} else {
p.closePrice = p.broker.Bid("") // Ask because we are long so we have to sell.
}
p.broker.Cash += p.Value() // Return the value of the position to the broker.
p.broker.SignalEmit("PositionClosed", p)
return nil
}
func (p *TestPosition) Closed() bool {
return p.closed
}
func (p *TestPosition) EntryPrice() float64 {
return p.entryPrice
}
func (p *TestPosition) ClosePrice() float64 {
return p.closePrice
}
func (p *TestPosition) EntryValue() float64 {
return p.entryPrice * p.units
}
func (p *TestPosition) Id() string {
return p.id
}
func (p *TestPosition) Leverage() float64 {
return p.leverage
}
func (p *TestPosition) PL() float64 {
return p.Value() - p.EntryValue()
}
func (p *TestPosition) Symbol() string {
return p.symbol
}
func (p *TestPosition) StopLoss() float64 {
return p.stopLoss
}
func (p *TestPosition) TakeProfit() float64 {
return p.takeProfit
}
func (p *TestPosition) Time() time.Time {
return p.time
}
func (p *TestPosition) Units() float64 {
return p.units
}
func (p *TestPosition) Value() float64 {
if p.closed {
return p.closePrice * p.units
}
var price float64
if p.units < 0 {
price = p.broker.Ask("")
} else {
price = p.broker.Bid("")
}
return price * p.units
}
type TestOrder struct {
id string
leverage float64
position *TestPosition
price float64
symbol string
stopLoss float64
takeProfit float64
time time.Time
orderType OrderType
units float64
}
func (o *TestOrder) Cancel() error {
return ErrCancelFailed
}
func (o *TestOrder) Fulfilled() bool {
return o.position != nil
}
func (o *TestOrder) Id() string {
return o.id
}
func (o *TestOrder) Leverage() float64 {
return o.leverage
}
func (o *TestOrder) Position() Position {
return o.position
}
func (o *TestOrder) Price() float64 {
return o.price
}
func (o *TestOrder) Symbol() string {
return o.symbol
}
func (o *TestOrder) StopLoss() float64 {
return o.stopLoss
}
func (o *TestOrder) TakeProfit() float64 {
return o.takeProfit
}
func (o *TestOrder) Time() time.Time {
return o.time
}
func (o *TestOrder) Type() OrderType {
return o.orderType
}
func (o *TestOrder) Units() float64 {
return o.units
}