Multi-Factor Long/Short Equity¶
The bread-and-butter of systematic equity: don't bet on one signal, combine several. We z-score value, quality, and momentum, add them into a single composite score, and build a dollar-neutral long/short book — long the best names, short the worst. Diversifying across factors smooths the ride when any one of them is out of favor.
In [1]:
try:
import convexpi.lab # noqa
except ImportError:
import subprocess, sys
subprocess.run([sys.executable, "-m", "pip", "install", "-q", "convexpi-lab"])
%matplotlib inline
import numpy as np
import matplotlib.pyplot as plt
print("ready")
The idea¶
Each factor is already cross-sectionally standardized each day, so we can simply sum them into a
composite and sort on it. Value (val_bm) buys cheap, quality (qual_roe) buys profitable, momentum
(mom_12m) buys winners — three largely independent edges in one book.
In [2]:
import numpy as np
from convexpi.lab import Strategy
def _long_short(signal, frac=0.2):
"""Dollar-neutral long/short: long the top `frac`, short the bottom `frac`, equal-weighted."""
s = np.nan_to_num(np.asarray(signal, dtype=float))
n = len(s); k = max(1, int(n * frac))
order = np.argsort(s)
w = np.zeros(n); w[order[-k:]] = 1.0 / k; w[order[:k]] = -1.0 / k
return w
class MyStrategy(Strategy):
"""Composite of value + quality + momentum, dollar-neutral top/bottom quintile."""
weights = {"val_bm": 1.0, "qual_roe": 1.0, "mom_12m": 1.0}
def on_day(self, day, features, prices, portfolio):
n = len(prices)
combo = np.zeros(n)
for name, wt in self.weights.items():
combo += wt * np.nan_to_num(features.get(name, np.zeros(n)))
return _long_short(combo, frac=0.2)
Out-of-sample evaluation¶
Train on the first half of a synthetic market, evaluate on the held-out second half — the same discipline as the leaderboard.
In [3]:
from convexpi.lab import SyntheticMarket, Grader
market = SyntheticMarket(n_stocks=80, n_days=1800, seed=1)
report = Grader(market).evaluate(MyStrategy())
print(f"in-sample Sharpe : {report.is_sharpe:+.2f}")
print(f"out-of-sample Sharpe: {report.oos_sharpe:+.2f}")
print(f"overfitting ratio : {report.overfitting_ratio:+.2f}")
oos = report.oos_result.daily_returns
fig, ax = plt.subplots(figsize=(8, 3))
ax.plot(np.cumprod(1 + oos)); ax.set_title("Out-of-sample equity curve"); ax.set_ylabel("growth of $1")
plt.tight_layout(); plt.show()
# How much does each factor contribute? Compare the composite to each factor alone.
single = {}
for f in ["val_bm", "qual_roe", "mom_12m"]:
class _S(MyStrategy):
weights = {f: 1.0}
single[f] = Grader(market).evaluate(_S()).oos_sharpe
print("OOS Sharpe — single factors:", {k: round(v, 2) for k, v in single.items()})
print("OOS Sharpe — composite :", round(report.oos_sharpe, 2))
print("Combining factors smooths in-sample — but beating OOS is hard; watch the overfitting ratio.")
What I'd try next¶
- Weight the factors by recent IC instead of equally.
- Neutralize the book to size or sector before sorting.
- Add a volatility scale so the book targets constant risk.
- Watch the overfitting ratio as you add factors — more knobs = more ways to fool yourself.