BatchBacktesting

1. Overview

Why it exists. A batch backtest is a small data pipeline, and the project keeps every step visible in Python instead of behind a dashboard.

Best fit. Comparing one strategy across many instruments with the same recorded assumptions, then reading the per-instrument and aggregate CSVs side by side.

3. Getting started

The README links a published example rather than a one-command install; the runner is the entry point.

• Runner: src/BatchBacktesting.py — process_instrument selects a stock or crypto price adapter, cleans the data and constructs a Backtest.
• Strategies: src/strategies.py.
• Analysis helpers: src/analyse.py.
• Published AAPL EMA chart (2018-04-04 to 2023-04-03): output/charts/EMA.

4. Features and components

5. How the strategies decide

EMA crossover. The default strategy computes exponential moving averages of the close, EMAₜ = α · Pₜ + (1 − α) · EMAₜ₋₁ with α = 2 / (n + 1): 0.095 for the 20-period average and 0.025 for the 80-period one, so the short average reacts about four times faster. On each bar:

• when EMA 20 crosses above EMA 80, the strategy closes any position and buys, with a stop at 90% and a target at 125% of the close;
• when EMA 80 crosses above EMA 20, it closes and sells short, with a stop at 110% and a target at 75%.

The 150-period average is computed but never used.

What the exits imply. With a stop 10% away and a target 25% away, a long trade risks 10 to make 25. Ignoring fees, it breaks even if more than 10 / (10 + 25) ≈ 28.6% of trades reach the target before the stop. Each round trip also pays the 0.2% commission twice, which raises that threshold slightly. Because the next crossover closes the position first, many trades end on a crossover rather than on the stop or the target.

flowchart TD
  B["New bar"] --> C1{"EMA20 crosses above EMA80?"}
  C1 -->|yes| L["Close position, buy: SL 0.90 × close, TP 1.25 × close"]
  C1 -->|no| C2{"EMA80 crosses above EMA20?"}
  C2 -->|yes| S["Close position, sell short: SL 1.10 × close, TP 0.75 × close"]
  C2 -->|no| H["Hold; the broker checks SL and TP"]

The other strategies. src/strategies.py defines twenty-three indicator strategies (RSI, Bollinger Bands, MACD, linear regression, ADX, OBV, Aroon, DEMA, ALMA, SMA …). Most follow the same pattern, and fourteen of them use the same 90/125% and 110/75% exits. Two of them do not test what their names suggest. In RSI, rsi30 and rsi70 are RSIs computed over 30 and 70 bars, not the 30 and 70 levels, so the strategy trades crossings between three RSI curves. BollingerBands uses only the first column of pandas_ta.bbands, the lower band.

Aggregation. process_instrument tags each Backtest.run() result with the instrument and the strategy, the runner concatenates them into one CSV per strategy, and analyse.py ranks instruments. get_max_return groups by instrument and takes max() column by column, so the “best” row mixes the best return with the best value of every other column.

6. Architecture

flowchart TD
  I["Instrument list and strategy"] --> P["ThreadPoolExecutor"]
  P --> D["Fetch and clean historical data"]
  D --> B["Backtest per instrument"]
  B --> R["Result with Instrument and Strategy"]
  R --> C["Per-instrument CSV"]
  R --> A["Aggregate strategy CSV"]
  B --> H["Optional HTML chart"]

7. Constraints and checks

• A failed instrument returns None; if every instrument fails, the later concatenation has no frames.
• groupby("Instrument").max() in analyse.py takes each column’s maximum independently, so a summary row can mix metrics from different runs. Select a real row by return before quoting its drawdown.
• The defaults do not cover slippage, short availability, delisted securities or corporate actions.
• Worker counts and dates above are code defaults and example dates, not measured throughput or a fixed range.

8. Links

• Repository
• Published example
• Workshop1, the beginner notebooks that lead into this batch runner

flowchart LR
  repo[("GitHub repository")]
  local["Local machine<br/>Python"]
  repo -->|"clone and run"| local