package factorzoo import ( "context" "fmt" "github.com/sajari/regression" "github.com/sirupsen/logrus" "github.com/c9s/bbgo/pkg/bbgo" "github.com/c9s/bbgo/pkg/fixedpoint" "github.com/c9s/bbgo/pkg/types" ) const ID = "factorzoo" var three = fixedpoint.NewFromInt(3) var log = logrus.WithField("strategy", ID) func init() { bbgo.RegisterStrategy(ID, &Strategy{}) } type IntervalWindowSetting struct { types.IntervalWindow } type Strategy struct { Symbol string `json:"symbol"` Market types.Market Interval types.Interval `json:"interval"` Quantity fixedpoint.Value `json:"quantity"` Position *types.Position `json:"position,omitempty"` activeMakerOrders *bbgo.LocalActiveOrderBook orderStore *bbgo.OrderStore tradeCollector *bbgo.TradeCollector session *bbgo.ExchangeSession book *types.StreamOrderBook prevClose fixedpoint.Value pvDivergenceSetting *IntervalWindowSetting `json:"pvDivergence"` pvDivergence *Correlation Ret []float64 Alpha [][]float64 T int64 prevER fixedpoint.Value } func (s *Strategy) ID() string { return ID } func (s *Strategy) Subscribe(session *bbgo.ExchangeSession) { log.Infof("subscribe %s", s.Symbol) session.Subscribe(types.KLineChannel, s.Symbol, types.SubscribeOptions{Interval: s.Interval}) } func (s *Strategy) ClosePosition(ctx context.Context, percentage fixedpoint.Value) error { base := s.Position.GetBase() if base.IsZero() { return fmt.Errorf("no opened %s position", s.Position.Symbol) } // make it negative quantity := base.Mul(percentage).Abs() side := types.SideTypeBuy if base.Sign() > 0 { side = types.SideTypeSell } if quantity.Compare(s.Market.MinQuantity) < 0 { return fmt.Errorf("order quantity %v is too small, less than %v", quantity, s.Market.MinQuantity) } submitOrder := types.SubmitOrder{ Symbol: s.Symbol, Side: side, Type: types.OrderTypeMarket, Quantity: quantity, Market: s.Market, } // s.Notify("Submitting %s %s order to close position by %v", s.Symbol, side.String(), percentage, submitOrder) createdOrders, err := s.session.Exchange.SubmitOrders(ctx, submitOrder) if err != nil { log.WithError(err).Errorf("can not place position close order") } s.orderStore.Add(createdOrders...) s.activeMakerOrders.Add(createdOrders...) return err } func (s *Strategy) placeOrders(ctx context.Context, orderExecutor bbgo.OrderExecutor, er fixedpoint.Value) { // if s.prevER.Sign() < 0 && er.Sign() > 0 { if er.Sign() >= 0 { submitOrder := types.SubmitOrder{ Symbol: s.Symbol, Side: types.SideTypeBuy, Type: types.OrderTypeMarket, Quantity: s.Quantity, // er.Abs().Mul(fixedpoint.NewFromInt(20)), } createdOrders, err := orderExecutor.SubmitOrders(ctx, submitOrder) if err != nil { log.WithError(err).Errorf("can not place orders") } s.orderStore.Add(createdOrders...) s.activeMakerOrders.Add(createdOrders...) // } else if s.prevER.Sign() > 0 && er.Sign() < 0 { } else { submitOrder := types.SubmitOrder{ Symbol: s.Symbol, Side: types.SideTypeSell, Type: types.OrderTypeMarket, Quantity: s.Quantity, // er.Abs().Mul(fixedpoint.NewFromInt(20)), } createdOrders, err := orderExecutor.SubmitOrders(ctx, submitOrder) if err != nil { log.WithError(err).Errorf("can not place orders") } s.orderStore.Add(createdOrders...) s.activeMakerOrders.Add(createdOrders...) } s.prevER = er } func (s *Strategy) Run(ctx context.Context, orderExecutor bbgo.OrderExecutor, session *bbgo.ExchangeSession) error { // initial required information s.session = session s.prevClose = fixedpoint.Zero // first we need to get market data store(cached market data) from the exchange session st, _ := session.MarketDataStore(s.Symbol) // setup the time frame size iw := types.IntervalWindow{Window: 50, Interval: s.Interval} // construct CORR indicator s.pvDivergence = &Correlation{IntervalWindow: iw} // bind indicator to the data store, so that our callback could be triggered s.pvDivergence.Bind(st) // s.pvDivergence.OnUpdate(func(corr float64) { // //fmt.Printf("now we've got corr: %f\n", corr) // }) s.Alpha = [][]float64{{}, {}, {}, {}, {}} s.Ret = []float64{} // thetas := []float64{0, 0, 0, 0} preCompute := 0 s.activeMakerOrders = bbgo.NewLocalActiveOrderBook(s.Symbol) s.activeMakerOrders.BindStream(session.UserDataStream) s.orderStore = bbgo.NewOrderStore(s.Symbol) s.orderStore.BindStream(session.UserDataStream) if s.Position == nil { s.Position = types.NewPositionFromMarket(s.Market) } s.tradeCollector = bbgo.NewTradeCollector(s.Symbol, s.Position, s.orderStore) s.tradeCollector.BindStream(session.UserDataStream) session.UserDataStream.OnStart(func() { log.Infof("connected") }) session.MarketDataStream.OnKLineClosed(func(kline types.KLine) { if kline.Symbol != s.Symbol || kline.Interval != s.Interval { return } if err := s.activeMakerOrders.GracefulCancel(ctx, s.session.Exchange); err != nil { log.WithError(err).Errorf("graceful cancel order error") } // amplitude volume divergence corr := fixedpoint.NewFromFloat(s.pvDivergence.Last()).Neg() // price mean reversion rev := fixedpoint.NewFromInt(1).Div(kline.Close) // alpha150 from GTJA's 191 paper a150 := kline.High.Add(kline.Low).Add(kline.Close).Div(three).Mul(kline.Volume) // momentum from WQ's 101 paper mom := fixedpoint.One.Sub(kline.Open.Div(kline.Close)).Mul(fixedpoint.NegOne) // opening gap ogap := kline.Open.Div(s.prevClose) log.Infof("corr: %f, rev: %f, a150: %f, mom: %f, ogap: %f", corr.Float64(), rev.Float64(), a150.Float64(), mom.Float64(), ogap.Float64()) s.Alpha[0] = append(s.Alpha[0], corr.Float64()) s.Alpha[1] = append(s.Alpha[1], rev.Float64()) s.Alpha[2] = append(s.Alpha[2], a150.Float64()) s.Alpha[3] = append(s.Alpha[3], mom.Float64()) s.Alpha[4] = append(s.Alpha[4], ogap.Float64()) // s.Alpha[5] = append(s.Alpha[4], 1.0) // constant ret := kline.Close.Sub(s.prevClose).Div(s.prevClose).Float64() s.Ret = append(s.Ret, ret) log.Infof("Current Return: %f", s.Ret[len(s.Ret)-1]) // accumulate enough data for cross-sectional regression, not time-series regression s.T = 20 if preCompute < int(s.T)+1 { preCompute++ } else { s.ClosePosition(ctx, fixedpoint.One) s.tradeCollector.Process() // rolling regression for last 20 interval alphas r := new(regression.Regression) r.SetObserved("Return Rate Per Timeframe") r.SetVar(0, "Corr") r.SetVar(1, "Rev") r.SetVar(2, "A150") r.SetVar(3, "Mom") r.SetVar(4, "OGap") var rdp regression.DataPoints for i := 1; i <= int(s.T); i++ { // alphas[t-1], previous alphas, dot not take current alpha into account, will cause look-ahead bias as := []float64{s.Alpha[0][len(s.Alpha[0])-(i+2)], s.Alpha[1][len(s.Alpha[1])-(i+2)], s.Alpha[2][len(s.Alpha[2])-(i+2)], s.Alpha[3][len(s.Alpha[3])-(i+2)], s.Alpha[4][len(s.Alpha[4])-(i+2)]} // alphas[t], current return rate rt := s.Ret[len(s.Ret)-(i+1)] rdp = append(rdp, regression.DataPoint(rt, as)) } r.Train(rdp...) r.Run() fmt.Printf("Regression formula:\n%v\n", r.Formula) // prediction := r.Coeff(0)*corr.Float64() + r.Coeff(1)*rev.Float64() + r.Coeff(2)*factorzoo.Float64() + r.Coeff(3)*mom.Float64() + r.Coeff(4) prediction, _ := r.Predict([]float64{corr.Float64(), rev.Float64(), a150.Float64(), mom.Float64(), ogap.Float64()}) log.Infof("Predicted Return: %f", prediction) s.placeOrders(ctx, orderExecutor, fixedpoint.NewFromFloat(prediction)) s.tradeCollector.Process() } s.prevClose = kline.Close }) return nil }