There is an ambiguous line between what can be confidently inferred from a backtest and what is simply a reflection of random market movements. The strength of a trend is best identified through various stress testing methods applied across broad sets of data.
The vision for this repository is to store a comprehensive annotated collection of trading systems, backtesting frameworks, strategy assessments, market theories, and ideas coded in Python Jupyter Notebook. Chapters in this repository cover various concepts, utilize multiple backtesting methods/libraries, and employ several ways to validate testing. Most trading systems are devised only from price/volume data indicators which are quantifiable, replicable, and can be measured.
"Backtesting"
Backtesting is a technique used in trading and investing to assess the performance of a trading strategy or investment approach using historical market data. By applying specific predetermined rules and parameters, backtesting can offer traders valuable insights into profitability potentials, risks and constrains, and perfomance comparisons to alternative strategies.
Why code it?
Algorithmic trading (and backtesting) offers speed, precision, and consistency beyond human capabilities. Code is an invaluable tool for defining and combining indicators into easily adjustable entry signals and creating live visualization comparison of performance metrics/benchmarks.
Tip
Find highlighted comments in every project for quick summary of concepts and analysis. Color scales indicate key findings, limitations, and improvements.
Strategy-Backtest-Commentary
001-MA Crossover______Simple trend-following optimization aimed to reduce drawdowns tested on 30-year SPX daily data
002-Random Entry______With robust risk-management and position sizing even random entries can be profitable (multi asset)
003-London Breakout___Designed to capitalize on the high liquidity and volatility of the FOREX market during the London session
004-
| Concept | Status | Description, notes |
|---|---|---|
| Moving average optimization | 001-MA Crossover | Can we validate entry parameters by optimizing across every relevant performance metric-theory, limitations |
| Mutli-asset backtest | Backtest trading a universe of assets, 1% of equity, sl - Idea; OR (if not taking signals but holding and rebalancing) rank strenght of signal and buy one or few top ranked | |
| Random Entry | Random entry & direction ... is it possible to beat the market with only good risk management (position sizing, atr sl) | |
| Quantitative Momentum | The Quantitative Momentum Investing Philosophy by Jack Vogel, Ph.D. - ranks stocks by momentum and trend strenght, rebalanced quaterly | |
| Year High or 100 Day High | Enter position (from a universe of assets) if price reaches yearly high (250/100/other period lookback), sell at 10% gain | |
| Break Out Indicator | Detect price break-outs by identifying trading range break-outs in combination with liquidity sweeps and n. of bars above/below MA | |
| Backtestesting tool | Documentation | Example | Description |
|---|---|---|---|
| NumPy (daily returns) | NumPy guide | 001-MA Crossover | Storing data in arrays/matrices - calculating daily returns, benchmark drawdowns, strategy performance |
| backtesting.py | kernc.github.io | 001-MA Crossover | Popular Python framework for inferring viability of trading strategies on historical data |
| Backtrader | backtrader.com | Write and reusable trading strategies, indicators, performance visualization | |
| VectorBT | vectorbt.pro | Ability to combine multiple strategy instances into a single multi-dimensional array, enabling highly efficient data processing | |
| zipline | zipline-trader | Backtesting/trading program compatible with Interactive Brokers and Alpaca | |
| build your own? | Reliability, control, scale, independence, Possible starting point;algotrading101blog | ||
| Testing method | Documentation? | Example | Description |
|---|---|---|---|
| Single-run automated | 001-MA Crossover | python arrays/matrices - calculating daily returns, benchmark drawdowns, strategy performance | |
| Multi-run optimization | 001-MA Crossover | Cross-parameter backtesting - allows for many tests with many entry signal combinations | |
| Walk-forward | Finding optimal in-sample trading parameters and checking the performance in the following time period for out-of-sample results | ||
| Monte Carlo simulations | Helps assess strategy's robustness by randomizing simulation parameters & inputs (trade sequence, skip n trades) | ||
| combination ? | Examine how the performance/robustness of a strategy changes across assets |
| Indicator | Example | Description | |
|---|---|---|---|
| Volume | |||
| Moving Average | 001-MA Crossover | Many traders use moving averages as the basis for a trend-following trading system as a confirmation signal to another indicator | |
| RSI | |||
| (A)TR | |||
| Market Open | |||
| MACD | |||
| WVAP | |||
| Bollinger Bands | |||
| PSAR | |||
| Beta | |||
Market Efficiency Theory states that market prices tend to be perfectly reflective of all information in the market, implying that asset prices are always trading at their fair value. Thus, under a 'random walk' assumption, historical market and volume data have no value in predicting future stock prices. In other words, 'technical analysis' is useless and trying to time the market is a loser's game.
Adaptive Market view suggests market efficiency isn't fixed, but it evolves with competition, conditions, and who controls the volume.
The movement of market prices is not always completely random
Investor psychology, time-varying investor preferences, over-reaction, under-reaction, transaction costs, informational constraints, and even widespread use of similar trading systems contribute to this nonrandomness
Trading systems can be developed to effectively exploit deviations from the random walk. Mechanical trading systems can be profitable, in part, because they are immune to greed and fear
Over-reaction is more prevalent for some types assets and time frames (suitable for mean-reversion systems), and under-reaction is more prevalent in others (momentum systems)
There are different ways to characterize risk, including volatility measured by standard deviation and the probability of experiencing a drawdown of a given size
Important
This repository is in early stage of production, contribution etc. to be added (open to improvement ideas, feedback, contributions)
Though most work is authentic, some studies employ open-source code by online traders (all scripts are credited and accompanied by author's original license)
*create seperate file/folder to be designated for resources, learning paths, tools, references
"Absorb what is useful, discard what is not, add what is uniquely your own"
Bruce Lee
"As far as we can discern, the sole purpose of human existence is to kindle a light in the darkness of mere being"
Carl Jung
Wholeness is not something you create, it is something you notice. It's the quiet realization that nothing is actually missing, now.
add:
resources, learning tools references, credits
hide unecessary outputs in .ipynb strategies
revisit- Algotrading: Position size β devils game
credits/resrouces
Youtube - The art of trading
Youtube - Benjamin
Youtube - Quantconnect
Rapusa blog
Youtube - Moon Dev
Algotrading101 blog
Alpha Vantage
Youtube β CodeTrading
Youtube β Martin Bell
Youtube - Neuralnine
Youtube - Chad Thrackray
Youtube - Lit Nomad