Cost-ignoring bias

Also seen as: Cost blindness, gross-only reporting

Which costs does a cost-blind backtest leave out?

Five separate charges are all missing from a gross figure, and they are missing for different reasons:

• Commissions and fees — a per-trade or per-value charge paid on entry and again on exit; see transaction cost for how the charge itself is defined.
• The bid-ask spread — you buy at the higher quote and sell at the lower one, so a round trip pays the spread even if the price never moves.
• Slippage — the gap between the price an order is quoted at and the price it actually fills at, which widens for larger orders and thinner instruments.
• Financing and interest — the cost of borrowed money and the markups applied to interest rates, which apply to leveraged and short positions.
• Tax — withholding on dividends and tax on realized capital gains, which depends on your tax regime.

The first three are paid per trade. That is what makes turnover, not cost per trade, the dominant term.

How does turnover multiply a small cost into a large one?

The drag a strategy pays each year is its all-in cost per round trip multiplied by how many round trips it makes — so a cost too small to notice on one trade becomes a large annual number for a strategy that trades often.

where $r_{\text{gross}}$ is the annual return before costs, $r_{\text{net}}$ is the annual return after them, $T$ is annual turnover (how many times a year the portfolio's value is fully traded), and $c$ is the all-in cost of one round trip as a fraction of the value traded. The relation is an approximation — it ignores compounding within the year and any tax — but it is the right first estimate, and it shows the shape of the problem: halving $c$ helps once, while halving $T$ helps every year.

A rebalance is the main source of turnover. A strategy that rebalances annually has $T$ near 1; one that rebalances fortnightly has $T$ in the twenties.

Worked example: a gross winner that is a net loser

A strategy rotates its whole portfolio roughly every two weeks — about 24 full rotations a year, so $T = 24$. Its all-in round-trip cost is 0.30% of the value traded, combining commission, spread, and slippage. Before costs it compounds at +6.0% a year.

The gross curve nearly doubles the starting capital while the net curve ends below it — a difference of about $9,000 on a $10,000 start, produced entirely by a cost of three tenths of one percent per round trip. Tax would land on top of this for a strategy that did finish in profit, since it applies to realized gains and a strategy that rotates fortnightly realizes them constantly.

Cut $T$ from 24 to 2 and the same 0.30% cost produces a 0.6% drag instead of 7.2%, turning the same gross 6.0% into roughly 5.4% net. Nothing about the strategy's ideas changed — only how often it acted on them.

Defaults in Fincanva

• Costs and tax are two of the three simulation assumptions applied to a backtest: Costs & interests covers trading costs, financing, and interest, and Taxes covers tax on dividends and realized gains.
• By default a result is shown with Costs & interests and Taxes off, so the figures you first see are gross of both — a cost-blind view by construction, which is why it is worth knowing which view you are reading.
• Switching Costs & interests on applies modelled trading costs including slippage to every fill; with it off, modelled costs and slippage are zero.
• The costs actually paid appear in the result's P&L breakdown, so the drag is readable as an amount rather than inferred from the gap between two curves.

How do you read the gap between the gross and net figures?

The gap is a measure of how much the conclusion depends on the cost assumptions rather than on the strategy. A strategy whose gross and net figures sit close together is largely insensitive to those assumptions, so its result stands or falls on its own rules; one whose figures diverge sharply is telling you that most of the gross result is being consumed on the way through, and that its conclusion changes with any change to turnover, instrument liquidity, or fee schedule.

Reading only the gross figure is a metric-level cherry-pick, and reaching for the gross figure because the net one contradicts an idea you already hold is confirmation bias. Neither is a statement about which figure is correct — both are correct answers to different questions. See the nine biases Fincanva helps you avoid for the product behavior in context.

Backtests show what would have happened — not what will. Fincanva provides no financial advice — see Is this financial advice?.