Pair Trading Strategy
Purpose
Select two highly correlated instruments (such as stocks from the same industry or BTC/ETH), monitor how far their price ratio (or spread) deviates from the mean, and trade against extreme deviations while waiting for mean reversion.
Signal Logic
- Compute the price ratio:
ratio = close_A / close_B
- Rolling mean and standard deviation:
mean = ratio.rolling(lookback).mean(), std = ratio.rolling(lookback).std()
- Z-score:
z = (ratio - mean) / std
- Signal generation:
- Z < -entry_z → long A, short B (ratio is too low, expected to revert)
- Z > +entry_z → short A, long B (ratio is too high, expected to revert)
- |Z| < exit_z → close the position (reverted back near the mean)
Implementation Notes
- Pair trading requires exactly two instruments (
codes array length = 2)
- The first instrument is A (
leg1), and the second is B (leg2)
- Signals for A and B are opposite: when A is long, B is short, and vice versa
- Equal-weight allocation only: A and B each take 50% of capital, with no precise hedge-ratio calculation
Parameters
| Parameter |
Default |
Description |
| lookback |
60 |
Lookback window for mean and standard deviation |
| entry_z |
2.0 |
Entry Z-score threshold |
| exit_z |
0.5 |
Exit Z-score threshold |
Example config.json
{
"source": "tushare",
"codes": ["601318.SH", "601628.SH"],
"start_date": "2023-01-01",
"end_date": "2024-12-31",
"initial_cash": 1000000,
"commission": 0.001,
"extra_fields": null
}
Cryptocurrency version:
{
"source": "okx",
"codes": ["BTC-USDT", "ETH-USDT"],
"start_date": "2024-01-01",
"end_date": "2024-12-31",
"initial_cash": 1000000,
"commission": 0.001,
"extra_fields": null
}
Common Pitfalls
codes must contain exactly 2 instruments, no more and no less- The date indexes of the two instruments must be aligned (use an inner join), otherwise the ratio calculation will be wrong
- Before the lookback window is filled, Z-scores are
NaN, so fill signals with 0
- Do not generate same-direction signals for both A and B; pair trading is fundamentally a long-short hedge
Dependencies
pip install pandas numpy
Signal Convention
- Instrument A:
0.5 = long, -0.5 = short, 0 = flat
- Instrument B: direction is opposite to A
1---
2name: pair-trading
3description: Pair trading strategy. Trades mean reversion using the spread/ratio Z-score of two correlated instruments. Requires at least two instruments.
4category: strategy
5---
6
7# Pair Trading Strategy
8
9## Purpose
10
11Select two highly correlated instruments (such as stocks from the same industry or BTC/ETH), monitor how far their price ratio (or spread) deviates from the mean, and trade against extreme deviations while waiting for mean reversion.
12
13## Signal Logic
14
151. **Compute the price ratio**: `ratio = close_A / close_B`
162. **Rolling mean and standard deviation**: `mean = ratio.rolling(lookback).mean()`, `std = ratio.rolling(lookback).std()`
173. **Z-score**: `z = (ratio - mean) / std`
184. **Signal generation**:
19 - Z < -entry_z → long A, short B (ratio is too low, expected to revert)
20 - Z > +entry_z → short A, long B (ratio is too high, expected to revert)
21 - |Z| < exit_z → close the position (reverted back near the mean)
22
23## Implementation Notes
24
25- Pair trading requires **exactly two instruments** (`codes` array length = 2)
26- The first instrument is A (`leg1`), and the second is B (`leg2`)
27- Signals for A and B are opposite: when A is long, B is short, and vice versa
28- **Equal-weight allocation only**: A and B each take 50% of capital, with no precise hedge-ratio calculation
29
30## Parameters
31
32| Parameter | Default | Description |
33|------|--------|------|
34| lookback | 60 | Lookback window for mean and standard deviation |
35| entry_z | 2.0 | Entry Z-score threshold |
36| exit_z | 0.5 | Exit Z-score threshold |
37
38## Example `config.json`
39
40```json
41{
42 "source": "tushare",
43 "codes": ["601318.SH", "601628.SH"],
44 "start_date": "2023-01-01",
45 "end_date": "2024-12-31",
46 "initial_cash": 1000000,
47 "commission": 0.001,
48 "extra_fields": null
49}
50```
51
52Cryptocurrency version:
53```json
54{
55 "source": "okx",
56 "codes": ["BTC-USDT", "ETH-USDT"],
57 "start_date": "2024-01-01",
58 "end_date": "2024-12-31",
59 "initial_cash": 1000000,
60 "commission": 0.001,
61 "extra_fields": null
62}
63```
64
65## Common Pitfalls
66
67- `codes` must contain exactly 2 instruments, no more and no less
68- The date indexes of the two instruments must be aligned (use an inner join), otherwise the ratio calculation will be wrong
69- Before the lookback window is filled, Z-scores are `NaN`, so fill signals with 0
70- Do not generate same-direction signals for both A and B; pair trading is fundamentally a long-short hedge
71
72## Dependencies
73
74```bash
75pip install pandas numpy
76```
77
78## Signal Convention
79
80- Instrument A: `0.5` = long, `-0.5` = short, `0` = flat
81- Instrument B: direction is opposite to A
HKUDS