Modify an example strategy to trade on the futures market

magenta.grimer

Help me in modifying this strategy to use it in the futures markets, not the cryptofutures one.

import xarray as xr

import qnt.backtester as qnbt
import qnt.data as qndata
import numpy as np
import pandas as pd


def load_data(period):
    futures = qndata.futures.load_data(tail=period, assets=["F_O", "F_LN"])
    crypto = qndata.cryptofutures.load_data(tail=period)
    return {"futures": futures, "crypto": crypto}, futures.time.values


def build_data_for_one_step(data, max_date: np.datetime64, lookback_period: int):
    min_date = max_date - np.timedelta64(lookback_period, "D")
    return {
        "futures": data["futures"].sel(time=slice(min_date, max_date)),
        "crypto": data["crypto"].sel(time=slice(min_date, max_date)),
    }


def predict_weights(market_data):
    def get_ml_model():
        from sklearn.linear_model import RidgeClassifier
        model = RidgeClassifier(random_state=18)
        return model

    def get_features(data):
        def remove_trend(prices_pandas_):
            prices_pandas = prices_pandas_.copy(True)
            assets = prices_pandas.columns
            for asset in assets:
                prices_pandas[asset] = np.log(prices_pandas[asset])
            return prices_pandas

        price = data.sel(field="close").ffill('time').bfill('time').fillna(0) # fill NaN
        for_result = price.to_pandas()
        features_no_trend_df = remove_trend(for_result)
        return features_no_trend_df

    def get_target_classes(data):

        price_current = data.sel(field="close").dropna('time')
        price_future = price_current.shift(time=-1).dropna('time')

        class_positive = 1
        class_negative = 0

        target_is_price_up = xr.where(price_future > price_current, class_positive, class_negative)
        return target_is_price_up.to_pandas()

    futures = market_data["futures"].copy(True)
    crypto = market_data["crypto"].copy(True)

    asset_name_all = crypto.coords['asset'].values
    features_all_df = get_features(futures)
    target_all_df = get_target_classes(crypto)

    predict_weights_next_day_df = crypto.sel(field="close").isel(time=-1).to_pandas()

    for asset_name in asset_name_all:
        target_for_learn_df = target_all_df[asset_name]
        feature_for_learn_df = features_all_df[:-1] # last value reserved for prediction

        # align features and targets
        target_for_learn_df, feature_for_learn_df = target_for_learn_df.align(feature_for_learn_df, axis=0, join='inner')

        model = get_ml_model()

        try:
            model.fit(feature_for_learn_df.values, target_for_learn_df)

            feature_for_predict_df = features_all_df[-1:]

            predict = model.predict(feature_for_predict_df.values)
            predict_weights_next_day_df[asset_name] = predict
        except:
            logging.exception("model failed")
            # if there is exception, return zero values
            return xr.zeros_like(crypto.isel(field=0, time=0))


    return predict_weights_next_day_df.to_xarray()


weights = qnbt.backtest(
    competition_type="cryptofutures",
    load_data=load_data,
    lookback_period=18,
    start_date='2014-01-01',
    strategy=predict_weights,
    window=build_data_for_one_step,
    analyze=True,
    build_plots=True
)```

magenta.grimer

@magenta-grimer I tried to change some values, like:

competition_type="cryptofutures",

in

competition_type="futures",

and other lines of code before, but I get errors...

Also from the original strategy it seems that we use futures data to predict the crypto prices,
we could try using both other futures data or even cryoto data to predict futures prices and submit it to the future competition?

Thanks

jeppe_and

@magenta-grimer The problem is connected with the fact that the futures competition needs data since January 1st, 2006 (beginning of the in-sample period); the crypto contest needs data since January 1st, 2014.

As a result it is ok to use futures data for predicting the crypto futures data, as the latter has a shorter time series, but the length of the crypto futures data is too short for predicting all futures data.

support

Hello.

If you changed only this line:

competition_type="futures",

you, probably, got the error min() arg is an empty sequence

It means, that the backtester can't calculate statistics because you return the output with the asset 'BTC' and the futures dataset does not contain this asset. You need to return the output with an asset which the futures dataset contains.

This a working example (see the lines with ###) :

import xarray as xr

import qnt.backtester as qnbt
import qnt.data as qndata
import numpy as np
import pandas as pd


def load_data(period):
    futures = qndata.futures.load_data(tail=period, assets=["F_O", "F_LN"])
    crypto = qndata.cryptofutures.load_data(tail=period)
    return {"futures": futures, "crypto": crypto}, futures.time.values


def build_data_for_one_step(data, max_date: np.datetime64, lookback_period: int):
    min_date = max_date - np.timedelta64(lookback_period, "D")
    return {
        "futures": data["futures"].sel(time=slice(min_date, max_date)),
        "crypto": data["crypto"].sel(time=slice(min_date, max_date)),
    }


def predict_weights(market_data):
    def get_ml_model():
        from sklearn.linear_model import RidgeClassifier
        model = RidgeClassifier(random_state=18)
        return model

    def get_features(data):
        def remove_trend(prices_pandas_):
            prices_pandas = prices_pandas_.copy(True)
            assets = prices_pandas.columns
            for asset in assets:
                prices_pandas[asset] = np.log(prices_pandas[asset])
            return prices_pandas

        price = data.sel(field="close").ffill('time').bfill('time').fillna(0) # fill NaN
        for_result = price.to_pandas()
        features_no_trend_df = remove_trend(for_result)
        return features_no_trend_df

    def get_target_classes(data):

        price_current = data.sel(field="close").dropna('time')
        price_future = price_current.shift(time=-1).dropna('time')

        class_positive = 1
        class_negative = 0

        target_is_price_up = xr.where(price_future > price_current, class_positive, class_negative)
        return target_is_price_up.to_pandas()

    futures = market_data["futures"].copy(True)
    crypto = market_data["crypto"].copy(True)

    asset_name_all = crypto.coords['asset'].values
    features_all_df = get_features(futures)
    target_all_df = get_target_classes(crypto)

    predict_weights_next_day_df = crypto.sel(field="close").isel(time=-1).to_pandas()

    for asset_name in asset_name_all:
        target_for_learn_df = target_all_df[asset_name]
        feature_for_learn_df = features_all_df[:-1] # last value reserved for prediction

        # align features and targets
        target_for_learn_df, feature_for_learn_df = target_for_learn_df.align(feature_for_learn_df, axis=0, join='inner')

        model = get_ml_model()

        try:
            model.fit(feature_for_learn_df.values, target_for_learn_df)

            feature_for_predict_df = features_all_df[-1:]

            predict = model.predict(feature_for_predict_df.values)
            predict_weights_next_day_df[asset_name] = predict
        except:
            logging.exception("model failed")
            # if there is exception, return zero values
            return xr.zeros_like(crypto.isel(field=0, time=0))


    output = predict_weights_next_day_df.to_xarray() ### 
    output = output.assign_coords(asset=['F_DX'])    ###
    return output                                    ###
    


weights = qnbt.backtest(
    competition_type="futures",
    load_data=load_data,
    lookback_period=18,
    start_date='2014-01-01',
    strategy=predict_weights,
    window=build_data_for_one_step,
    analyze=True,
    build_plots=True
)

And yes, @jeppe_and said is right. It will be impossible to develop a futures strategy using the bitcoin market data because the bitcoin futures contracts and bitcoin itself are very young. The data time series starts from 2013. The in-sample period for the futures contest should start from 2006.

Regards.

magenta.grimer

This post is deleted!

magenta.grimer

@support said in Modify an example strategy to trade on the futures market:

Ok, is it possible to use a simple technical strategy for the dates before 1-1-2014 and then switch to the ML strategy? Can you help in coding it?

Thanks

support

@magenta-grimer

Hello. This is such an example:

import xarray as xr

import qnt.backtester as qnbt
import qnt.data as qndata
import numpy as np
import pandas as pd
import qnt.ta as qnta


def load_data(period):
    futures = qndata.futures.load_data(tail=period, assets=["F_NQ", "F_DX"])
    try:
        # this will fail when there is no data
        crypto = qndata.cryptofutures.load_data(tail=period, assets=["BTC"])
    except:
        crypto = None
    return {"futures": futures, "crypto": crypto}, futures.time.values


def build_data_for_one_step(data, max_date: np.datetime64, lookback_period: int):
    min_date = max_date - np.timedelta64(lookback_period, "D")
    futures = data["futures"].sel(time=slice(min_date, max_date)).copy(True)
    try:
        crypto = data["crypto"].sel(time=slice(min_date, max_date)).copy(True)
    except:
        crypto = None
    return {
        "futures": futures,
        "crypto": crypto,
    }


def predict_weights(market_data):
    def get_ml_model():
        from sklearn.linear_model import RidgeClassifier
        model = RidgeClassifier(random_state=18)
        return model

    def get_features(data):
        def remove_trend(prices_pandas_):
            prices_pandas = prices_pandas_.copy(True)
            assets = prices_pandas.columns
            for asset in assets:
                prices_pandas[asset] = np.log(prices_pandas[asset])
            return prices_pandas

        price = data.sel(field="close").ffill('time').bfill('time').fillna(0) # fill NaN
        for_result = price.to_pandas()
        features_no_trend_df = remove_trend(for_result)
        return features_no_trend_df

    def get_target_classes(data):

        price_current = data.sel(field="close").dropna('time')
        price_future = price_current.shift(time=-1).dropna('time')

        class_positive = 1
        class_negative = 0

        target_is_price_up = xr.where(price_future > price_current, class_positive, class_negative)
        return target_is_price_up.to_pandas()

    futures = market_data["futures"]
    crypto = market_data["crypto"]

    asset_name_all = futures.asset.values

    weights_df = futures.sel(field="close").isel(time=-1).to_pandas()

    for asset_name in asset_name_all:
        
        features_data = xr.concat([crypto, futures.sel(asset=[asset_name])], dim='asset')
        features_all_df = get_features(features_data)
        target_all_df = get_target_classes(futures.sel(asset=[asset_name]))
        
        target_for_learn_df = target_all_df[asset_name]
        feature_for_learn_df = features_all_df[:-1] # last value reserved for prediction

        # align features and targets
        target_for_learn_df, feature_for_learn_df = target_for_learn_df.align(feature_for_learn_df, axis=0, join='inner')

        model = get_ml_model()

        try:
            model.fit(feature_for_learn_df.values, target_for_learn_df)

            feature_for_predict_df = features_all_df[-1:]

            predict = model.predict(feature_for_predict_df.values)
            weights_df[asset_name] = predict
        except:
            logging.exception("model failed")
            # if there is exception, return zero values
            return xr.zeros_like(crypto.isel(field=0, time=0))

    output = weights_df.to_xarray()
    output = output.drop(['time','field'], errors='ignore')
    return output


def ta_strategy(market_data):
    close = market_data["futures"].sel(field='close') 
    roc = qnta.roc(close, 20) 
    output = xr.where(roc > 0, 1, 0).isel(time=-1)
    output = output.drop(['time','field'], errors='ignore')
    return output
    

def hybrid_strategy(data):
#     try:
#         return predict_weights(data)
#     except:
#         return ta_strategy(data)
    if data['futures'].time[-1] >= pd.to_datetime('2014-01-01'):
        return predict_weights(data)
    else:
        return ta_strategy(data)


weights = qnbt.backtest(
    competition_type="futures",
    load_data=load_data,
    lookback_period=60,
    start_date='2006-01-01',
    strategy=hybrid_strategy,
    window=build_data_for_one_step,
    analyze=True,
    build_plots=True
)

magenta.grimer

@support Thank you very much!

I hope to don't disturb you so often in the future @support , but before that I need help to "get going".