A transaction cost is a charge incurred when buying or selling a financial asset. In algorithmic trading, it converts a paper gain into a more realistic result by reducing returns for each executed transaction. section 11.3.1
The cost matters when comparing a strategy with simply buying and holding an asset. A backtest can subtract a fixed charge, such as $1, for every recorded buy or sell; frequent signals therefore face a larger cost burden. section 11.3.1
Use transaction costs to:
- Reduce summed trade cash movements by the fee multiplied by the number of transactions. section 11.3.1
- Set a more demanding benchmark for whether a trading strategy improves on holding shares from the start to the end of a period. section 11.3.1
- Distinguish ETF expense ratios from the transaction fees avoided by purchasing a diversified fund instead of many underlying stocks separately. section 1.2.3
Applying costs in a backtest
A Python backtest can accept transaction_costs = 1, calculate a strategy’s gain, and then subtract transaction_costs multiplied by the number of recorded cash movements. The resulting gain reflects trading friction rather than an idealized signal-only return. section 11.3.1
Listing 11.2 Executing a single backtest
def run_single_backtest(ticker: str, cash, start_date: str, end_date: str,
transaction_costs = 1) -> pd.DataFrame:
df = load_data(ticker, start_date, end_date)
shares = round(cash / df["Close"][0])
print(f"Shares: {shares}. start price: {df["Close"][0]}. end price: "
f"{df["Close"].iloc[-1]}. to beat: {df["Close"].iloc[-1]*shares –
df["Close"][0]*shares –
transaction_costs}")
df = add_sma(df, 10, 20)
res = backtest(df, 10, 20, shares, True)
gain = res['cash_movement'].sum()
gain = gain - transaction_costs * len(res['cash_movement'])
print(f"Gains: {gain})")
return df
results_nvda
= run_single_backtest("NVDA", 1000, "2015-01-01", "2025-01-01")
Inspect the single-backtest implementation that loads prices, determines share quantity, records trades, and deducts transaction costs from the reported gain. section 11.3.1
