Backtesting

Backtesing framework written in C++17 to simulate trading strategy for BTC/USD pair.

DISCLAIMER ⚠️🛑

Warning

THIS FRAMEWORK IS NOT INTENDED FOR ANY REAL-WORLD INVESTMENT DECISIONS. THE USE OF THIS FRAMEWORK AND THE DATA PROVIDED IS AT YOUR OWN RISK. I DO NOT TAKE ANY RESPONSIBILITY FOR THE CORRECTNESS OF THE DATA OR ANY OUTPUTS GENERATED BY THIS FRAMEWORK WHEN APPLIED IN REAL-WORLD SCENARIOS. THIS FRAMEWORK IS DESIGNED SOLELY FOR LEARNING AND EDUCATIONAL PURPOSES.

Table of Contents

• Dependency
• Project Structure
• Data Download
• Compiling & Building
• Tick Data Generation
• Strategy Example And Other Terms
• How to Run
• How To Add New Strategy
• Performance Details
• Memory Allocation Test
• Task & Improvements
• License

Dependency

common-util added as git submodule. gnu plot On mac easy installation would be brew install gnuplot googletest for unit test cases. C++17, CMake, Clang,

Project-Structure

├── backtesting
│   ├── base (This compiles as static lib which get used in both data generation and simulation)
│   │   ├── account (Keep track of account balance/simulate buy and sell)
│   │   ├── common_interface
│   │   ├── price_history (OHLC data clean up)
│   │   ├── trade_simulator (Interface to add new trade strategy)
│   │   └── util (Read/Write into binary file, getting cmd line argument)
│   ├── execution (actually running the simulation)
│   ├── logs (log generation which get used to plot a graph)
│   └── simulators (Trading strategy and Trade Simulator logic)
│       └── strategy
├── data (Create this folder to hold ohlc data)
├── data_generator (Logic to convert TPV to OHLC and change frequency of OHLC)
├── external (Added dependancy as git submodule, memory map file, logger, time, command line argument and string formate util)
├── quick_run (bash script to quickly running the project)
└── result_plot (logic to plot a graph after running simulation)

Data-Download

Time Price Volume (TPV) data is hosted on kaggle. It can be downloaded directly or with use of wget link to download.

Create a folder named data and unzip data there with the name bitstamp_tick_data.csv

Compiling-Building

1. clone the repo git clone https://github.com/xpd54/backtesting
2. update submodule git submodule update --init --recursive
3. create a build folder and build mkdir build && cd build && cmake .. && make

This generate 3 exicutables

1. ohlc_generator (To convert TPV to binary form of OHLC data formate)
2. trade_simulator (Execute trade simulation)
3. plot (plot graph with evaluation log)

Tick-Data-Generation

Any trade simulation we would like to run on different frequency of tick data. ohlc_generator does this to do it quickly with [start_time - end_time) (time range to consider for ohlc (open high low close))

cd quick_run && chmod +x quick_ohlc_generation.sh ./quick_ohlc_generation.sh (might take a minute or two as tpv data is close to 3.2gb). I suggest checking quick_ohlc_generation.sh before running it, naming convention is self explanatory.

Strategy-and-Terms

Rebalancing Trade Strategy:- The basic idea here is to keep the portfolio value (BTC * BTC price + USD) in a way where cryptocurrency weight is defined by alpha and it's allowed to go up and down by epsilon. Example:- alpha 0.7 means that we would like to keep 70% of the total portfolio value in the base (crypto) currency, and 30% in the quote currency (USD) for whole execution time. So Current quick run between [start_time='2017-01-01' - end_time='2024-01-01'] we would maintain 70% in BTC and 30% in USD for the whole 7 years while looking to make a profit.

epsilon 0.05f means the maximum allowed deviation from the desired alpha allocation. We allow the actual base (crypto) currency allocation to be within the interval: [alpha x (1 - epsilon) - alpha x (1 + epsilon)].

So [0.735 - 0.665] which means it's allowed to allocate 73.5% to 66.5% in BTC and the rest corresponding in USD. codebase is self-explanatory but will add more clarity in words soon!!

How-to-Run

use quick_run/quick_simulation.sh to run trade simulation over 5 min frequency data. Example commands are HERE

./trade_simulator \
--input_price_history_binary_file="../data/bitstamp_tick_data_5min.mov" \ (change it to 30min.mov and 1h.mov for other frequency)
--output_account_log_file="../data/account.log" \
--output_simulator_log_file="../data/simulator.log" \
--start_time="2017-01-01" \ (start execution from jan 2017 to jan 2024)
--end_time="2024-01-01" \
--start_base_balance=1.0 \ (we have 1 btc to start with)
--start_quote_balance=0.0 \ (we have 0 usd to start with)

Here we can see with the rebalancing trade strategy we have a gain of 2043.60%. But if we would have just bought and hold it would have been 4272.85%. Well HODL strategy is the best in the crypto market do time travel and buy in 2011 and sell in 2024 that's HODL.

[2025-02-03 20:47:00.665] [0x7ff8599feb40] [INFO] rebalancing_trade_simulator[0.7000|0.0500] evaluation
[2025-02-03 20:47:05.889] [0x7ff8599feb40] [INFO] --------------- Time Period ---------------  Strategy gain | Base gain(HODL) | Score | volatility
[2025-02-03 20:47:05.890] [0x7ff8599feb40] [INFO] [2017-01-01 00:00:00 - 2024-01-01 00:00:00):  2043.6094%  | 4272.8594%  | 0.4902  | 0.0000  | 0.0000
[2025-02-03 20:47:05.895] [0x7ff8599feb40] [INFO] Evaluated in 0:0:5sec

if we run the same strategy over diff frequencies it should be close enough (2055% gain) on 1h frequency with a similar 0.49 score.

[2025-02-03 20:55:00.884] [0x7ff8599feb40] [INFO] rebalancing_trade_simulator[0.7000|0.0500] evaluation
[2025-02-03 20:55:01.650] [0x7ff8599feb40] [INFO] --------------- Time Period ---------------  Strategy gain | Base gain(HODL) | Score | volatility
[2025-02-03 20:55:01.650] [0x7ff8599feb40] [INFO] [2017-01-01 00:00:00 - 2024-01-01 00:00:00):  2055.8738%  | 4271.8188%  | 0.4931  | 0.0000  | 0.0000
[2025-02-03 20:55:01.657] [0x7ff8599feb40] [INFO] Evaluated in 0:0:1sec

We should look into our evaluation log which shows we are selling our 0.3 BTC immediately where our alpha was set for 0.7 (70% in BTC) we achieved that balance.

1483228800,966.3400,966.9900,964.6000,966.6000,102.4848,1.0000,0.0000,0.0000,,,,,
1483232400,966.6000,966.6000,962.5400,963.8700,149.0256,1.0000,0.0000,0.0000,,,,,
1483232400,966.6000,966.6000,962.5400,963.8700,149.0256,0.7000,287.9200,1.4500,MARKET,SELL,0.300000,,
1483236000,964.3500,965.7500,961.9900,963.9700,94.2674,0.7000,287.9200,1.4500,,,,,

If we plot the portfolio value we see it started with 1 BTC and goes down (get sold and allocate in USD) asap after that it trades in the range of [0.735 - 0.665] (73.5% to 66.5%) of allocation.

./plot --output_account_log_file="../data/account.log"

(The number of data points that I am using is limited to 30000 as higher data point plot generation becomes slower)

Let's run it with multiple combinations of alpha and epsilon [alpha|epsilon]. Again if we look int the score we can see highest score suggest allocating most of the assets in BTC which is HODL (buy and hold).

./trade_simulator \
--input_price_history_binary_file="../data/bitstamp_tick_data_1h.mov" \
--output_account_log_file="../data/account.log" \
--output_simulator_log_file="../data/simulator.log" \
--evaluation_period_months=6 \
--start_time="2017-01-01" \
--end_time="2024-01-01" \
--start_base_balance=1.0 \
--start_quote_balance=0.0 \
--evaluate_combination=1

[2025-02-03 21:05:18.790] [0x7ff8599feb40] [INFO] Evaluation Combination of simulators
[2025-02-03 21:05:18.886] [0x7ff8599feb40] [INFO] rebalancing_trade_simulator[0.9000|0.2000]: 1.0000
[2025-02-03 21:05:18.886] [0x7ff8599feb40] [INFO] rebalancing_trade_simulator[0.9000|0.1000]: 0.9948
[2025-02-03 21:05:18.886] [0x7ff8599feb40] [INFO] rebalancing_trade_simulator[0.9000|0.0500]: 0.9905
[2025-02-03 21:05:18.886] [0x7ff8599feb40] [INFO] rebalancing_trade_simulator[0.9000|0.0100]: 0.9812
[2025-02-03 21:05:18.886] [0x7ff8599feb40] [INFO] rebalancing_trade_simulator[0.7000|0.2000]: 0.9586
[2025-02-03 21:05:18.886] [0x7ff8599feb40] [INFO] rebalancing_trade_simulator[0.7000|0.0500]: 0.9509
[2025-02-03 21:05:18.886] [0x7ff8599feb40] [INFO] rebalancing_trade_simulator[0.7000|0.1000]: 0.9483
[2025-02-03 21:05:18.886] [0x7ff8599feb40] [INFO] rebalancing_trade_simulator[0.7000|0.0100]: 0.9421
[2025-02-03 21:05:18.886] [0x7ff8599feb40] [INFO] rebalancing_trade_simulator[0.5000|0.1000]: 0.9132
[2025-02-03 21:05:18.886] [0x7ff8599feb40] [INFO] rebalancing_trade_simulator[0.5000|0.2000]: 0.9126
[2025-02-03 21:05:18.886] [0x7ff8599feb40] [INFO] rebalancing_trade_simulator[0.5000|0.0500]: 0.9085
[2025-02-03 21:05:18.886] [0x7ff8599feb40] [INFO] rebalancing_trade_simulator[0.5000|0.0100]: 0.8923
[2025-02-03 21:05:18.886] [0x7ff8599feb40] [INFO] rebalancing_trade_simulator[0.3000|0.2000]: 0.8647
[2025-02-03 21:05:18.886] [0x7ff8599feb40] [INFO] rebalancing_trade_simulator[0.3000|0.1000]: 0.8625
[2025-02-03 21:05:18.886] [0x7ff8599feb40] [INFO] rebalancing_trade_simulator[0.3000|0.0500]: 0.8556
[2025-02-03 21:05:18.886] [0x7ff8599feb40] [INFO] rebalancing_trade_simulator[0.3000|0.0100]: 0.8429
[2025-02-03 21:05:18.886] [0x7ff8599feb40] [INFO] rebalancing_trade_simulator[0.1000|0.2000]: 0.8064
[2025-02-03 21:05:18.886] [0x7ff8599feb40] [INFO] rebalancing_trade_simulator[0.1000|0.1000]: 0.8048
[2025-02-03 21:05:18.886] [0x7ff8599feb40] [INFO] rebalancing_trade_simulator[0.1000|0.0500]: 0.8024
[2025-02-03 21:05:18.886] [0x7ff8599feb40] [INFO] rebalancing_trade_simulator[0.1000|0.0100]: 0.7963

_{TODO:- Add more details on how rebalancing trade strategy works}

How-To-Add-New-Strategy

By implementation of an interface provided into backtesting/base/trade_simulator/trade_simulator.hpp And updating backtesting/simulators/simulator_factory.hpp and backtesting/simulators/strategy/common_strategy_config.hpp it's easily can be achieved.

Performance-Details

I am using mmap to read and write into the file. Which map the disk file directly to memory. I wrote a common-util which have easy to use header-only lib.

Memory-Allocation-Test

Memory Leaks Output

Process:         trade_simulator [71547]
Path:            /Users/USER/*/trade_simulator
Load Address:    0x102783000
Identifier:      trade_simulator
Version:         0
Code Type:       X86-64
Platform:        macOS
Parent Process:  leaks [71546]

Date/Time:       2025-02-09 17:44:26.653 +0800
Launch Time:     2025-02-09 17:44:15.726 +0800
OS Version:      macOS 15.1.1 (24B91)
Report Version:  7
Analysis Tool:   /usr/bin/leaks

Physical footprint:         1840K
Physical footprint (peak):  46.6M
Idle exit:                  untracked
----

leaks Report Version: 4.0, multi-line stacks
Process 71547: 224 nodes malloced for 55 KB
Process 71547: 0 leaks for 0 total leaked bytes.

Task-And-Improvements

• Single instance Logger for both console and logfile.
• Use CRTP to avoid using virtual.
• Improve read/write with memory mapped file.
• Write OHLC data to a binary file.
• Run account logging on different processes.
• Rebalancing trade strategy doc.
• Financial indicators
• Integrate google test
• Unit Test cases
• Check and Fix memory leaks.