Back to Community
Replication of backtest results in Research environment

Using the research environmment, I'm able to backtest strategies with longer computation time. While this is great, I have a list of similar strategies that are able to be backtested in the regular Q backtester. To compare strategies, i will need to bridge the gap between calculating performance numbers that Q backtester does versus in the research environment such that, if I carry out the same strategy on both research and backtest environment, the statistics returned should be virtually the same. However, i can't get my results to match.

Let's say I want to execute a min. var strategy on both the backtester and research environment. My goal is to have the ending results match and let's assume my only concern statistic is Max Drawdown
attached below is the research notebook AND the backtest strategy. The code i'm using to calculate max_drawdown is said to be the same as the one Q uses taken from zipline/finance/risk/period.py

Is there anything I'm missing out on?

Loading notebook preview...
Notebook previews are currently unavailable.
3 responses

Attached in this post is the Q backtest, as you can see the max drawdown is 31.3% versus 38.86% on the research environment.

Clone Algorithm
7
Loading...
Backtest from to with initial capital
Total Returns
--
Alpha
--
Beta
--
Sharpe
--
Sortino
--
Max Drawdown
--
Benchmark Returns
--
Volatility
--
Returns 1 Month 3 Month 6 Month 12 Month
Alpha 1 Month 3 Month 6 Month 12 Month
Beta 1 Month 3 Month 6 Month 12 Month
Sharpe 1 Month 3 Month 6 Month 12 Month
Sortino 1 Month 3 Month 6 Month 12 Month
Volatility 1 Month 3 Month 6 Month 12 Month
Max Drawdown 1 Month 3 Month 6 Month 12 Month
import pandas as pd
import numpy as np
from scipy import optimize
from sklearn.covariance import OAS, LedoitWolf
import statsmodels.regression.linear_model as lm
import math

def initialize(context):
    schedule_function(trade, 
                      date_rule=date_rules.week_end(), 
                      time_rule=time_rules.market_close(minutes=1), 
                      half_days=False)
    context.lookback = 252*2 # 252*n = n years
    set_commission(commission.PerShare(cost=0.00))
    set_slippage(slippage.VolumeShareSlippage(volume_limit=1, price_impact=0))
    context.stocks = symbols('XLF', 'XLE', 'XLU', 'XLK', 'XLB', 'XLP', 'XLY', 'XLI', 'XLV')
    context.firstime = True
 
def trade(context,data):
        w = estimate(context,data)
    
        # Close all current positions
        for stock in context.portfolio.positions:
            if stock not in w.index:
               order_target(stock, 0)  

       #Order
        for i,n in enumerate(w):
            if w.index[i] in data:
                if w.index[i] in context.portfolio.positions: #If position already exists
                    order_target_percent(w.index[i], n)
                else:
                    order_percent(w.index[i], n)
    
def estimate(context,data):
        # Structure a portfolio such that it maximizes dividend yield while lowering volatility
        # Covar estimation
        price_history = history(bar_count=context.lookback, frequency="1d", field='price')
        ret = (price_history/price_history.shift(5)-1).dropna(axis=0)[5::5]
        covar = ret.cov() # Normal Cov
        
        w = min_var_weights(covar, context.stocks)

        # LOG EVERYTHING
        esigma = np.sqrt(np.dot(w.T,np.dot(covar,w)))*math.sqrt(52)*100
        log.info('Expected Volatility: '+str(esigma)+'%')
        w = pd.Series(np.round(1000*w)/1000)
        log.info('Number of Stocks: '+str(sum(abs(w)>0)))
        log.info('Current Leverage: '+str(context.account.leverage))
        w.index = ret.columns
        log.info(w)
        return w  #Cash Cushion of X%

def handle_data(context, data):
    pass

def min_var_weights(cov, stocks):
    '''
    Returns a dictionary of sid:weight pairs.
    '''
    cov = pd.DataFrame(2*cov)
    x = np.array([0.]*(len(cov)+1))
    #x = np.ones(len(cov) + 1)
    x[-1] = 1.0
    p = lagrangize(cov)
    weights = np.linalg.solve(p, x)[:-1]
    weights = pd.Series(weights, index=stocks)
    return weights


def lagrangize(df):
    '''
    Utility funcion to format a DataFrame 
    in order to solve a Lagrangian sysem. 
    '''
    df = df
    df['lambda'] = np.ones(len(df))
    z = np.ones(len(df) + 1)
    x = np.ones(len(df) + 1)
    z[-1] = 0.0
    x[-1] = 1.0
    m = [i for i in df.as_matrix()]
    m.append(z)    
    return pd.DataFrame(np.array(m))
There was a runtime error.

Is there a reason you can't use perf_manual max drawdown key?

perf_manual = algo_obj.run(data.transpose(2,1,0))  
print perf_manual["max_drawdown"]  
Disclaimer

The material on this website is provided for informational purposes only and does not constitute an offer to sell, a solicitation to buy, or a recommendation or endorsement for any security or strategy, nor does it constitute an offer to provide investment advisory services by Quantopian. In addition, the material offers no opinion with respect to the suitability of any security or specific investment. No information contained herein should be regarded as a suggestion to engage in or refrain from any investment-related course of action as none of Quantopian nor any of its affiliates is undertaking to provide investment advice, act as an adviser to any plan or entity subject to the Employee Retirement Income Security Act of 1974, as amended, individual retirement account or individual retirement annuity, or give advice in a fiduciary capacity with respect to the materials presented herein. If you are an individual retirement or other investor, contact your financial advisor or other fiduciary unrelated to Quantopian about whether any given investment idea, strategy, product or service described herein may be appropriate for your circumstances. All investments involve risk, including loss of principal. Quantopian makes no guarantees as to the accuracy or completeness of the views expressed in the website. The views are subject to change, and may have become unreliable for various reasons, including changes in market conditions or economic circumstances.

Even then, the values are off from the actual backtest, which gives one number, and assuming we use the last value of max_drawdown, it's still off.