QAStats.

What it is

QAStats is a Python library for financial reporting on portfolios and strategies, built on pandas. It starts from empyrical and pyfolio, in their -reloaded versions, and adapts them to an in-house workflow.

It was built to fix a problem of method. Without a common tool, each team member computed and presented results in their own way, and reports could not be compared. QAStats sets a single way of measuring: the same statistics and the same charts for every strategy.

The input is simple: the daily percentage change of the strategy and of its benchmark. It does not work with individual trades, but with the end-of-day aggregate. It can be used as is, but it is meant as a base: understand how it works inside and adapt it to each case.

What's included

The qastats package (version 3.0.1) has these modules:

• stats: daily (get_daily_stats) and monthly statistics tables, drawdowns in relevant periods (get_drawdown_stats), daily VAMI (get_vami_diario), worst drawdown periods, SQN and omega ratios, and dollar VaR (var_usd).
• plots: cumulative returns against the benchmark, heatmap and distribution of monthly returns, annual returns, underwater drawdown and drawdown periods, stress events, rolling volatility, alpha, beta, Sharpe and VaR, and Monte Carlo.
• timeseries: maximum drawdown and top drawdowns, rolling metrics and a bootstrap forecast cone.
• randomizer: parametric Monte Carlo of the equity curve, with return and maximum drawdown percentiles.
• reports: create_return_dd, the full report in a single figure. It is saved as PDF (noname.pdf when no name is given).
• interesting_periods: market events used by the relevant-period drawdowns.
• extension and utils: extend_pandas (adds varusd to pandas objects) and table and date utilities.

The repository ships an example, examples/example1.ipynb, with its return series (examples/diario_normal.pickle), and the report it produces (noname.pdf, noname.png).

Version 3 adds the SQN ratio to the advanced statistics, updates the list of relevant events, fixes bugs and installs with pip.

Installation and first run

In a dedicated conda environment, clone the repository and install the dependencies and the package. requirements.txt includes a patched version of quantstats that the library imports.

conda create -n qastats python
conda activate qastats
git clone https://github.com/quantarmyz/QAStats
cd QAStats
pip install -r requirements.txt
pip install .

The repository example uses a DataFrame with two columns of daily returns: Close (the strategy) and bench (the benchmark). In the example the benchmark is synthetic: returns of the same strategy drawn at random.

import numpy as np
import pandas as pd
import qastats

data = pd.read_pickle('examples/diario_normal.pickle')
data['bench'] = np.random.choice(data['Close'], len(data))

qastats.stats.get_daily_stats(data)

The full report takes a single call:

qastats.reports.create_return_dd(data)

How it fits

QAStats is the last piece of the lab chain. Data comes from the Datalake, the backtest runs on Zipline in Quantstack and the report comes from QAStats.

In the quant stack of the reports, pyfolio and empyrical handle reporting. QAStats was built to bring pyfolio's ideas into an in-house format, on top of empyrical's calculations. It is not part of the Quantstack image: install it separately with the steps above.