Note from Towards Data Science’s editors: While we allow independent authors to publish articles in accordance with our rules and guidelines, we do not endorse each author’s contribution. You should not rely on an author’s works without seeking professional advice. See our Reader Terms for details.

Towards Data Science编辑的注意事项： 尽管我们允许独立作者按照我们的 规则和指南 发表文章 ，但我们不认可每位作者的贡献。 您不应在未征求专业意见的情况下依赖作者的作品。 有关 详细信息， 请参见我们的 阅读器条款 。

算法交易策略开发 (Algorithmic Trading Strategy Development)

Backtesting is the hallmark of quantitative trading. Backtesting takes historical or synthetic market data and tests the profitability of algorithmic trading strategies. This topic demands expertise in statistics, computer science, mathematics, finance, and economics. This is exactly why in large quantitative trading firms there are specific roles for individuals with immense knowledge (usually at the doctorate level) of the respective subjects. The necessity for expertise cannot be understated as it separates winning (or seemingly winning) trading strategies from losers. My goal for this article is to break up what I consider an elementary backtesting process into a few different sections…

回测是定量交易的标志。 回测采用历史或合成市场数据，并测试算法交易策略的盈利能力。 本主题需要统计，计算机科学，数学，金融和经济学方面的专业知识。 这就是为什么在大型定量交易公司中，对于具有相应学科的丰富知识(通常是博士学位)的个人，要发挥特定的作用。 专业知识的必要性不可低估，因为它将成功(或看似成功)的交易策略与失败者分开。 本文的目的是将我认为是基本的回测过程分为几个不同的部分……

• I. The Backtesting Engine

  I.回测引擎

• II. Historical and Synthetic Data Management

  二。 历史和综合数据管理

• III. Backtesting Metrics

  三， 回测指标

• IV. Live Implementation

  IV。 现场实施

• V. Pitfalls in Strategy Development

  五 ，战略发展中的陷阱

回测引擎 (The Backtesting Engine)

The main backtesting engine will be built in Python using a library called backtrader. Backtrader makes it incredibly easy to build trading strategies and implement them on historical data immediately. The entire library centers around the Cerebro class. As the name implies, you can think of this as the brain or engine of the backtest.

主要的回测引擎将使用称为backtrader的库在Python中构建 。 Backtrader使得建立交易策略并立即在历史数据上实施交易变得异常容易。 整个图书馆围绕着Cerebro类。 顾名思义，您可以将其视为回测的大脑或引擎。

import backtrader as btdef backtesting_engine(symbol, strategy, fromdate, todate, args=None):"""Primary function for backtesting, not entirely parameterized"""# Backtesting Enginecerebro = bt.Cerebro()

I decided to build a function so I can parameterize different aspects of the backtest on the fly (and eventually build a pipeline). The next backtrader class we need to discuss is the Strategy class. If you look at the function I built to house the instance of Cerebro you’ll see that it takes a strategy as an input — this is expecting a backtrader Strategy class or subclass. Subclasses of the Strategy class are exactly what we will be using to build our own strategies. If you need a polymorphism & inheritance refresher or don’t know what that means see my article 3 Python Concepts in under 3 minutes. Let’s build a strategy for our backtest…

我决定构建一个函数，以便可以动态地对回测的不同方面进行参数化(并最终构建管道)。 我们需要讨论的下一个回购者类是策略类。 如果您查看我为容纳Cerebro实例而构建的功能，您会看到它以策略作为输入-这是期待的回购商Strategy类或子类。 策略类的子类正是我们用来构建自己的策略的类。 如果您需要多态和继承复习器，或者不知道这意味着什么，请在3分钟之内阅读我的文章3 Python概念 。 让我们为回测制定策略...

from models.isolation_model import IsolationModel
import backtrader as bt
import pandas as pd
import numpy as npclass MainStrategy(bt.Strategy):'''Explanation:MainStrategy houses the actions to take as a stream of data flows in a linear fashion with respect to time'''def log(self, txt, dt=None):''' Logging function fot this strategy'''dt = dt or self.datas[0].datetime.date(0)print('%s, %s' % (dt.isoformat(), txt))def __init__(self, data):# Keep a reference to the "close" line in the data[0] dataseriesself.dataopen = self.datas[0].openself.datahigh = self.datas[0].highself.datalow = self.datas[0].lowself.dataclose = self.datas[0].closeself.datavolume = self.datas[0].volume

In the MainStrategy, Strategy subclass above I create fields that we are interested in knowing about before making a trading decision. Assuming a daily data frequency the fields we have access to every step in time during the backstep are…

在上面的MainStrategy，Strategy子类中，我创建了我们感兴趣的字段，这些字段是我们在做出交易决策之前要了解的。 假设每天的数据频率，在后退步骤中我们可以访问的每个时间段的字段是……

• Day price open
  当日价格开盘
• Day price close
  当天收盘价
• Day price low
  当日价格低
• Day price high
  当天价格高
• Day volume
  日成交量

It doesn’t take a genius to understand that you cannot use the high, low, and close to make an initial or intraday trading decision as we don’t have access to that information in realtime. However, it is useful if you want to store it and access that information from previous days.

理解您不能使用高，低和接近来做出初始或盘中交易决策并不需要天才，因为我们无法实时访问该信息。 但是，如果您要存储它并访问前几天的信息，这将很有用。

The big question in your head right now should be where is the data coming from?

您现在脑海中最大的问题应该是数据来自哪里？

The answer to that question is in the structure of the backtrader library. Before we run the function housing Cerebro we will add all the strategies we want to backtest, and a data feed — the rest is all taken care of as the Strategy superclass holds the datas series housing all the market data.

该问题的答案在于backtrader库的结构。 在运行带有Cerebro的功能之前，我们将添加我们要回测的所有策略以及一个数据提要-其余所有工作都将得到处理，因为Strategy超类拥有容纳所有市场数据的数据系列。

This makes our lives extremely easy.

这使我们的生活变得极为轻松。

Let’s finish the MainStrategy class by making a simple long-only mean reversion style trading strategy. To access a tick of data we override the next function and add trading logic…

让我们通过制定一个简单的长期均值回归样式交易策略来结束MainStrategy类。 要访问数据变动，我们将覆盖下一个功能并添加交易逻辑…

from models.isolation_model import IsolationModel
import backtrader as bt
import pandas as pd
import numpy as npclass MainStrategy(bt.Strategy):'''Explanation:MainStrategy houses the actions to take as a stream of data flows in a linear fashion with respect to time
'''def log(self, txt, dt=None):''' Logging function fot this strategy'''dt = dt or self.datas[0].datetime.date(0)print('%s, %s' % (dt.isoformat(), txt))def __init__(self):self.dataclose = self.datas[0].closeself.mean = bt.indicators.SMA(period=20)self.std = bt.indicators.StandardDeviation()self.orderPosition = 0self.z_score = 0def next(self):self.log(self.dataclose[0])z_score = (self.dataclose[0] - self.mean[0])/self.std[0]if (z_score >= 2) & (self.orderPosition > 0):self.sell(size=1)if z_score <= -2:self.log('BUY CREATE, %.2f' % self.dataclose[0])self.buy(size=1)self.orderPosition += 1

Every day gets a z-score calculated for it (number of standard deviations from the mean) based on its magnitude we will buy or sell equity. Please note, yes I am using the day’s close to make the decision, but I’m also using the day’s close as an entry price to my trade — it would be wiser to use the next days open.

每天都会根据其大小来计算z得分(相对于平均值的标准偏差的数量)，我们将买卖股票。 请注意，是的，我使用当天的收盘价来做决定，但是我也将当天的收盘价用作我交易的入场价-使用第二天的开盘价会更明智。

The next step is to add the strategy to our function housing the Cerebro instance…

下一步是将策略添加到容纳Cerebro实例的功能中…

import backtrader as bt
import pyfolio as pfdef backtesting_engine(symbol, strategy, fromdate, todate, args=None):"""Primary function for backtesting, not entirely parameterized"""# Backtesting Enginecerebro = bt.Cerebro()# Add a Strategy if no Data Required for the modelif args is None:cerebro.addstrategy(strategy)# If the Strategy requires a Model and therefore dataelif args is not None:cerebro.addstrategy(strategy, args)

In the code above I add the strategy provided to the function to the Cerebro instance. This is beyond the scope of the article but I felt pressed to include it. If the strategy that we are backtesting requires additional data (some AI model) we can provide it in the args parameter and add it to the Strategy subclass.

在上面的代码中，我将为函数提供的策略添加到Cerebro实例。 这超出了本文的范围，但是我迫切希望将其包括在内。 如果要回测的策略需要其他数据(某些AI模型)，则可以在args参数中提供它，并将其添加到Strategy子类中。

Next, we need to find a source of historical or synthetic data.

接下来，我们需要找到历史或综合数据的来源。

历史和综合数据管理 (Historical and Synthetic Data Management)

I use historical and synthetic market data synonymously as some researchers have found synthetic data indistinguishable (and therefore is more abundant as its generative) from other market data. For the purposes of this example we will be using the YahooFinance data feed provided by backtrader, again making implementation incredibly easy…

我将历史和综合市场数据作为同义词使用，因为一些研究人员发现综合数据与其他市场数据是无法区分的(因此，其生成量更大)。 出于本示例的目的，我们将使用backtrader提供的YahooFinance数据供稿，再次使实现变得异常容易。

import backtrader as bt
import pyfolio as pfdef backtesting_engine(symbol, strategy, fromdate, todate, args=None):"""Primary function for backtesting, not entirely parameterized"""# Backtesting Enginecerebro = bt.Cerebro()# Add a Strategy if no Data Required for the modelif args is None:cerebro.addstrategy(strategy)# If the Strategy requires a Model and therefore dataelif args is not None:cerebro.addstrategy(strategy, args)# Retrieve Data from YahooFinancedata = bt.feeds.YahooFinanceData(dataname=symbol,fromdate=fromdate,  # datetime.date(2015, 1, 1)todate=todate,  # datetime.datetime(2016, 1, 1)reverse=False)# Add Data to Backtesting Enginecerebro.adddata(data)

It’s almost ridiculous how easy it is to add a data feed to our backtest. The data object uses backtraders YahooFinanceData class to retrieve data based on the symbol, fromdate, and todate. Afterwards, we simply add that data to the Cerebro instance. The true beauty in backtrader’s architecture comes in realizing that the data is automatically placed within, and iterated through every strategy added to the Cerebro instance.

将数据提要添加到我们的回测中是多么容易，这简直太荒谬了。 数据对象使用backtraders YahooFinanceData类根据符号，起始日期和起始日期检索数据。 然后，我们只需将该数据添加到Cerebro实例中即可。 backtrader架构的真正美在于实现将数据自动放入其中，并通过添加到Cerebro实例的每种策略进行迭代。

回测指标 (Backtesting Metrics)

There are a variety of metrics used to evaluate the performance of a trading strategy, we’ll cover a few in this section. Again, backtrader makes it extremely easy to add analyzers or performance metrics. First, let’s set an initial cash value for our Cerebro instance…

有多种指标可用于评估交易策略的绩效，本节将介绍其中的一些指标。 同样，backtrader使添加分析器或性能指标变得非常容易。 首先，让我们为Cerebro实例设置初始现金价值...

import backtrader as bt
import pyfolio as pfdef backtesting_engine(symbol, strategy, fromdate, todate, args=None):"""Primary function for backtesting, not entirely parameterized"""# Backtesting Enginecerebro = bt.Cerebro()# Add a Strategy if no Data Required for the modelif args is None:cerebro.addstrategy(strategy)# If the Strategy requires a Model and therefore dataelif args is not None:cerebro.addstrategy(strategy, args)# Retrieve Data from YahooFinancedata = bt.feeds.YahooFinanceData(dataname=symbol,fromdate=fromdate,  # datetime.date(2015, 1, 1)todate=todate,  # datetime.datetime(2016, 1, 1)reverse=False)# Add Data to Backtesting Enginecerebro.adddata(data)# Set Initial Portfolio Valuecerebro.broker.setcash(100000.0)

There are a variety of metrics we can use to evaluate risk, return, and performance — let’s look at a few of them…

我们可以使用多种指标来评估风险，回报和绩效-让我们看看其中的一些…

夏普比率 (Sharpe Ratio)

As referenced in the show Billions, the amount of return per unit of risk, the Sharpe Ratio. Defined as follows…

如显示的“十亿”所示，单位风险的收益金额即夏普比率。 定义如下...

Several downfalls, and assumptions. It’s based on historical performance, assumes normal distributions, and can often be used to incorrectly compare portfolios. However, it’s still a staple to any strategy or portfolio.

几次失败和假设。 它基于历史表现，假定为正态分布，并且经常被用来错误地比较投资组合。 但是，它仍然是任何策略或投资组合的主要内容。

System Quality Number

系统质量编号

An interesting metric I’ve always enjoyed using as it includes volume of transactions in a given period. Computed by multiplying the SQN by the number of trades and dividing the product by 100. Below is a chart of values across different segments, regions, and sectors.

我一直喜欢使用的有趣指标，因为它包含给定时间段内的交易量。 通过将SQN乘以交易数量并将乘积除以100进行计算。以下是不同细分市场，区域和部门的价值图表。

Let’s now add these metrics to Cerebro and run the backtest…

现在让我们将这些指标添加到Cerebro并进行回测...

import backtrader as bt
import pyfolio as pfdef backtesting_engine(symbol, strategy, fromdate, todate, args=None):"""Primary function for backtesting, not entirely parameterized"""# Backtesting Enginecerebro = bt.Cerebro()# Add a Strategy if no Data Required for the modelif args is None:cerebro.addstrategy(strategy)# If the Strategy requires a Model and therefore dataelif args is not None:cerebro.addstrategy(strategy, args)# Retrieve Data from Alpacadata = bt.feeds.YahooFinanceData(dataname=symbol,fromdate=fromdate,  # datetime.date(2015, 1, 1)todate=todate,  # datetime.datetime(2016, 1, 1)reverse=False)# Add Data to Backtesting Enginecerebro.adddata(data)# Set Initial Portfolio Valuecerebro.broker.setcash(100000.0)# Add Analysis Toolscerebro.addanalyzer(bt.analyzers.SharpeRatio, _name='sharpe')# Of course we want returns too ;Pcerebro.addanalyzer(bt.analyzers.Returns, _name='returns')cerebro.addanalyzer(bt.analyzers.SQN, _name='sqn')# Starting Portfolio Valueprint('Starting Portfolio Value: %.2f' % cerebro.broker.getvalue())# Run the Backtesting Enginebacktest = cerebro.run()# Print Analysis and Final Portfolio Valueprint('Final Portfolio Value: %.2f' % cerebro.broker.getvalue())print('Return: ', backtest[0].analyzers.returns.get_analysis())print('Sharpe Ratio: ', backtest[0].analyzers.sharpe.get_analysis())print('System Quality Number: ', backtest[0].analyzers.sqn.get_analysis())

Now, all we have to do is run the code above and we get the performance of our strategy based on the metrics we specified…

现在，我们要做的就是运行上面的代码，并根据指定的指标获得策略的性能……

from datetime import datetime
from strategies.reversion_strategy import LongOnlyReversion
from tools.backtesting_tools import backtesting_engine"""Script for backtesting strategies
"""if __name__ == '__main__':# Run backtesting enginebacktesting_engine(symbol='AAPL', strategy=LongOnlyReversion,fromdate=datetime(2018, 1, 1), todate=datetime(2019, 1, 1))

Here is the output…

这是输出...

Starting Portfolio Value: 100000.002018-01-30, 160.95
2018-01-30, BUY CREATE, 160.952018-01-31, 161.4
2018-02-01, 161.74
2018-02-02, 154.72
2018-02-02, BUY CREATE, 154.72
2018-02-05, 150.85
2018-02-05, BUY CREATE, 150.85
2018-02-06, 157.16
2018-02-07, 153.792018-02-08, 149.56
2018-02-09, 151.39
2018-02-12, 157.49
2018-02-13, 159.07
2018-02-14, 162.0
2018-02-15, 167.44
2018-02-16, 166.9
2018-02-20, 166.332018-02-21, 165.58
2018-02-22, 166.96
2018-02-23, 169.87
2018-02-26, 173.23
2018-02-27, 172.66
2018-02-28, 172.42018-03-01, 169.38
2018-03-02, 170.55
2018-03-05, 171.15
2018-03-06, 171.0
2018-03-07, 169.41
2018-03-08, 171.26
2018-03-09, 174.2
2018-03-12, 175.89
2018-03-13, 174.19
2018-03-14, 172.71
2018-03-15, 172.92
2018-03-16, 172.31
2018-03-19, 169.67
2018-03-20, 169.62
2018-03-21, 165.77
2018-03-21, BUY CREATE, 165.77
2018-03-22, 163.43
2018-03-22, BUY CREATE, 163.43
2018-03-23, 159.65
2018-03-23, BUY CREATE, 159.652018-03-26, 167.22
2018-03-27, 162.94
2018-03-28, 161.14
2018-03-29, 162.4
2018-04-02, 161.33
2018-04-03, 162.99
2018-04-04, 166.1
2018-04-05, 167.252018-04-06, 162.98
2018-04-09, 164.59
2018-04-10, 167.69
2018-04-11, 166.91
2018-04-12, 168.55
2018-04-13, 169.12
2018-04-16, 170.18
2018-04-17, 172.522018-04-18, 172.13
2018-04-19, 167.25
2018-04-20, 160.4
2018-04-23, 159.94
2018-04-24, 157.71
2018-04-25, 158.4
2018-04-26, 158.952018-04-27, 157.11
2018-04-30, 159.96
2018-05-01, 163.67
2018-05-02, 170.9
2018-05-03, 171.212018-05-04, 177.93
2018-05-07, 179.22
2018-05-08, 180.08
2018-05-09, 181.352018-05-10, 183.94
2018-05-11, 183.24
2018-05-14, 182.81
2018-05-15, 181.15
2018-05-16, 182.842018-05-17, 181.69
2018-05-18, 181.03
2018-05-21, 182.31
2018-05-22, 181.85
2018-05-23, 183.02
2018-05-24, 182.812018-05-25, 183.23
2018-05-29, 182.57
2018-05-30, 182.18
2018-05-31, 181.57
2018-06-01, 184.84
2018-06-04, 186.392018-06-05, 187.83
2018-06-06, 188.482018-06-07, 187.97
2018-06-08, 186.26
2018-06-11, 185.812018-06-12, 186.83
2018-06-13, 185.29
2018-06-14, 185.39
2018-06-15, 183.48
2018-06-18, 183.392018-06-19, 180.42
2018-06-20, 181.21
2018-06-21, 180.2
2018-06-22, 179.68
2018-06-25, 177.0
2018-06-25, BUY CREATE, 177.002018-06-26, 179.2
2018-06-27, 178.94
2018-06-28, 180.24
2018-06-29, 179.862018-07-02, 181.87
2018-07-03, 178.7
2018-07-05, 180.14
2018-07-06, 182.64
2018-07-09, 185.172018-07-10, 184.95
2018-07-11, 182.55
2018-07-12, 185.61
2018-07-13, 185.9
2018-07-16, 185.5
2018-07-17, 186.022018-07-18, 185.0
2018-07-19, 186.44
2018-07-20, 186.01
2018-07-23, 186.18
2018-07-24, 187.53
2018-07-25, 189.292018-07-26, 188.7
2018-07-27, 185.56
2018-07-30, 184.52
2018-07-31, 184.89
2018-08-01, 195.78
2018-08-02, 201.51
2018-08-03, 202.09
2018-08-06, 203.14
2018-08-07, 201.24
2018-08-08, 201.37
2018-08-09, 202.96
2018-08-10, 202.35
2018-08-13, 203.66
2018-08-14, 204.52
2018-08-15, 204.99
2018-08-16, 208.0
2018-08-17, 212.15
2018-08-20, 210.08
2018-08-21, 209.67
2018-08-22, 209.682018-08-23, 210.11
2018-08-24, 210.77
2018-08-27, 212.5
2018-08-28, 214.22
2018-08-29, 217.42
2018-08-30, 219.41
2018-08-31, 221.952018-09-04, 222.66
2018-09-05, 221.21
2018-09-06, 217.53
2018-09-07, 215.78
2018-09-10, 212.882018-09-11, 218.26
2018-09-12, 215.55
2018-09-13, 220.76
2018-09-14, 218.25
2018-09-17, 212.44
2018-09-18, 212.79
2018-09-19, 212.92
2018-09-20, 214.542018-09-21, 212.23
2018-09-24, 215.28
2018-09-25, 216.65
2018-09-26, 214.92
2018-09-27, 219.34
2018-09-28, 220.112018-10-01, 221.59
2018-10-02, 223.56
2018-10-03, 226.282018-10-04, 222.3
2018-10-05, 218.69
2018-10-08, 218.19
2018-10-09, 221.212018-10-10, 210.96
2018-10-11, 209.1
2018-10-12, 216.57
2018-10-15, 211.94
2018-10-16, 216.612018-10-17, 215.67
2018-10-18, 210.63
2018-10-19, 213.84
2018-10-22, 215.14
2018-10-23, 217.172018-10-24, 209.72
2018-10-25, 214.31
2018-10-26, 210.9
2018-10-29, 206.94
2018-10-30, 207.98
2018-10-31, 213.42018-11-01, 216.67
2018-11-02, 202.3
2018-11-02, BUY CREATE, 202.30
2018-11-05, 196.56
2018-11-05, BUY CREATE, 196.562018-11-06, 198.68
2018-11-06, BUY CREATE, 198.68
2018-11-07, 204.71
2018-11-08, 204.02018-11-09, 200.06
2018-11-12, 189.99
2018-11-12, BUY CREATE, 189.99
2018-11-13, 188.09
2018-11-13, BUY CREATE, 188.092018-11-14, 182.77
2018-11-14, BUY CREATE, 182.77
2018-11-15, 187.28
2018-11-16, 189.362018-11-19, 181.85
2018-11-20, 173.17
2018-11-20, BUY CREATE, 173.17
2018-11-21, 172.972018-11-23, 168.58
2018-11-26, 170.86
2018-11-27, 170.48
2018-11-28, 177.042018-11-29, 175.68
2018-11-30, 174.73
2018-12-03, 180.84
2018-12-04, 172.88
2018-12-06, 170.952018-12-07, 164.86
2018-12-10, 165.94
2018-12-11, 165.0
2018-12-12, 165.46
2018-12-13, 167.27
2018-12-14, 161.91
2018-12-17, 160.41
2018-12-18, 162.49
2018-12-19, 157.42
2018-12-20, 153.45
2018-12-20, BUY CREATE, 153.45
2018-12-21, 147.48
2018-12-21, BUY CREATE, 147.48
2018-12-24, 143.67
2018-12-24, BUY CREATE, 143.67
2018-12-26, 153.78
2018-12-27, 152.78
2018-12-28, 152.86
2018-12-31, 154.34
Final Portfolio Value: 100381.48
Return:  OrderedDict([('rtot', 0.0038075421029081335), ('ravg', 1.5169490449833201e-05), ('rnorm', 0.003830027474518453), ('rnorm100', 0.3830027474518453)])
Sharpe Ratio:  OrderedDict([('sharperatio', None)])
System Quality Number:  AutoOrderedDict([('sqn', 345.9089929588268), ('trades', 2)])

现场实施 (Live Implementation)

This is less of a section more of a collection of resources that I have for implementing trading strategies. If you want to move forward and implement your strategy in a live market check out these articles…

这不是本节的更多内容，更多的是我用于实施交易策略的资源集合。 如果您想继续前进并在实时市场中实施策略，请查看以下文章…

• Algorithmic Trading with Python

  使用Python进行算法交易

• Build an AI Stock Trading Bot for Free

  免费建立AI股票交易机器人

• Algorithmic Trading System Development

  算法交易系统开发

• Multithreading in Python for Finance

  Python for Finance中的多线程

• How to Build a Profitable Stock Trading Bot

  如何建立一个有利可图的股票交易机器人

• Build a Neural Network to Manage a Stock Portfolio

  建立一个神经网络来管理股票投资组合

算法交易策略的陷阱 (Algorithmic Trading Strategy Pitfalls)

• Not understanding the drivers of profitability in a strategy: There are some that believe, regardless of the financial or economic drivers, if you can get enough data and stir it up in such a way that you can develop a profitable trading strategy. I did too when I was younger, newer to the algorithmic trading, and more naive. The truth is if you can’t understand the nature of your own trading algorithmic it will never be profitable.

  不了解策略中获利能力的驱动因素：有人认为，无论财务或经济驱动因素如何，如果您能够获取足够的数据并以一种可以制定有利可图的交易策略的方式进行搅动，便会认为这是可行的。 当我还年轻，算法交易较新且天真时，我也这样做了。 事实是，如果您不了解自己的交易算法的性质，它将永远不会盈利。

• Not understanding how to split historical or synthetic data to both search for and optimize strategies: This one is big and the consequences are dire. I had a colleague back in college tell me he developed a system with a 300% return trading energy. Naturally, and immediately, I knew he completely overfitted a model to historical data and never actually implemented it in live data. Asking a simple question about his optimization process confirmed my suspicion.

  不了解如何拆分历史数据或合成数据以同时搜索和优化策略 ：这一步很大，后果非常严重。 我上大学时曾有一位同事告诉我，他开发了一种具有300％的收益交易能量的系统。 自然而又立即地，我知道他完全将模型拟合到历史数据中，而从未真正在实时数据中实现它。 问一个关于他的优化过程的简单问题证实了我的怀疑。

• Not willing to accept down days: This one is interesting, and seems to have a greater effect on newer traders more than experienced ones. The veterans know there will be down days, losses, but the overarching goal is to have (significantly) more wins than losses. This issue generally arises in deployment — assuming that the other processes

  不愿接受低迷的日子：这很有趣，而且似乎对新手的影响要大于经验丰富的人。 退伍军人知道会有损失的日子，但是总的目标是胜利(多于失败)。 通常在部署过程中会出现此问题-假设其他过程

• Not willing to retire a formerly profitable system: Loss control, please understand risk-management and loss control. Your system was successful, and that’s fantastic — that means you understand the design process and can do it again. Don’t get caught up in how well it did look at how well it’s doing. This has overlap with understanding the drivers of profitability in your strategy — you will know when and why it’s time to retire a system if you understand why it’s no longer profitable.

  不愿意淘汰以前有利可图的系统：损失控制，请了解风险管理和损失控制。 您的系统成功了，这太棒了-这意味着您了解设计过程并可以再次执行。 不要被它的表现看得如何。 这与了解战略中获利的驱动因素有很多重叠-如果您了解为什么不再有利润，您将知道何时以及为何该退休系统。