A minimal moving-average crossover backtest

Let us turn the SMA crossover idea into a small, self-contained vectorized backtest using just pandas and NumPy. This shows the logic; a real framework adds proper order handling and reporting.

The strategy

Go long when the fast moving average is above the slow one; be flat otherwise. A long-only, always-or-never position.

import numpy as np
import pandas as pd

def backtest_sma(df, fast=20, slow=50, cost_bps=2):
    px = df["close"]
    sma_fast = px.rolling(fast).mean()
    sma_slow = px.rolling(slow).mean()

    # Signal: 1 when fast above slow, else 0
    signal = (sma_fast > sma_slow).astype(int)

    # CRITICAL: trade on the NEXT bar to avoid look-ahead
    position = signal.shift(1).fillna(0)

    # Costs charged when the position changes
    trades = position.diff().abs().fillna(0)
    cost = trades * (cost_bps / 10000.0)

    ret = px.pct_change().fillna(0)
    strat_ret = position * ret - cost
    equity = (1 + strat_ret).cumprod()
    return equity, strat_ret

The single most important line

position = signal.shift(1).fillna(0)

The shift(1) is everything. The signal is computed from a bar's close, but you cannot act on it until the next bar opens. Without that shift you are trading on information you did not yet have — instant look-ahead bias and a backtest that looks magical and is worthless.

Reading the result

equity, r = backtest_sma(df)
sharpe = np.sqrt(252) * r.mean() / r.std()
print("final equity:", round(equity.iloc[-1], 3))
print("approx Sharpe:", round(sharpe, 2))

Honest caveats

This omits slippage beyond the flat cost, assumes you always fill at the close, ignores position sizing and shorting, and uses a single instrument and period. It is a teaching skeleton. Any equity curve it produces is hypothetical — treat it as a starting point for the rigorous testing in the rest of this track, not a result to trust.