量化回测精度影响

数据精度越高, 越能仿真实际情况, 尤其是对于高频策略, 日线无法捕捉日内波动.

股债金ETF混合均值回归策略, 用5m线回测:

Final Portfolio Value: 222353.54

Returns: OrderedDict({‘rtot’: 0.10595128430313513, ‘ravg’: 0.00043963188507524953, ‘rnorm’: 0.11715718966047985, ‘rnorm100’: 11.715718966047985})

Sharpe Ratio: OrderedDict({‘sharperatio’: 0.7643104578858475})

Draw Down: AutoOrderedDict({‘len’: 864, ‘drawdown’: 3.370534777068345, ‘moneydown’: 7755.919454999996, ‘max’: AutoOrderedDict({‘len’: 2518, ‘drawdown’: 5.9512311139187775, ‘moneydown’: 12782.288910690258})})

日线回测:

Final Portfolio Value: 224532.42

Returns: OrderedDict({‘rtot’: 0.11570275810353982, ‘ravg’: 0.0004820948254314159, ‘rnorm’: 0.12917569830963194, ‘rnorm100’: 12.917569830963194})

Sharpe Ratio: OrderedDict({‘sharperatio’: 0.7749162145844245})

Draw Down: AutoOrderedDict({‘len’: 18, ‘drawdown’: 3.3609850500329754, ‘moneydown’: 7808.95915499999, ‘max’: AutoOrderedDict({‘len’: 49, ‘drawdown’: 5.1694001517392065, ‘moneydown’: 11084.355427599541})})

从结果来看, 这个策略5m调仓还不如日调仓, 推测是交易频率过高, 进一步看了日志:

日线策略手续费合计: 64.77

5m线手续费合计: 69.05

这样看来, 就是atr_mult波动乘数还是设置的偏低了, 导致很多波动作为噪声被操作掉.

需要提高atr_mult. 有时间还是需要放到5年时间, 把组合全部用5m线再跑一边, 优化参数.