Transaction Costs in Trading: What Most Backtests Ignore

The most common failure mode in quantitative trading is not a bad model — it is a good model that ignores the cost of actually executing trades. Commissions, spreads, market impact, slippage, and opportunity cost form an invisible tax that can turn a profitable backtest into a money-losing strategy in production.

1. The Backtest-to-Live Performance Gap

Every practitioner encounters it eventually. A backtest shows a Sharpe ratio of 2.0, compounding at 25% per year. The strategy goes live. Six months later, realized performance is half of what was simulated — or worse. The culprit is almost always transaction costs. Backtests that assume frictionless execution are, at best, misleading. At worst, they create the illusion of alpha where none exists.

Transaction costs are not a single number. They are a composite of five distinct components, each with different drivers, magnitudes, and behaviors. Understanding each one is essential for building strategies that survive contact with real markets.

2. Component 1: Commissions

Commissions are the most visible cost and, ironically, the one that matters least for most retail traders today. In October 2019, Charles Schwab announced the elimination of commissions on online stock and ETF trades. Within days, TD Ameritrade, E*TRADE, and Fidelity followed suit. By the end of that month, zero-commission trading had become the standard across major U.S. retail brokers.

This was a watershed moment, but it is important to understand what “zero commission” actually means. Retail brokers still route orders through market makers (a practice called payment for order flow, or PFOF) and generate revenue from the spread between the price at which the market maker fills retail orders and the price available on lit exchanges. The commission has not disappeared — it has been embedded in execution quality.

For institutional traders, commissions remain a real cost. Typical institutional equity commissions range from 1 to 5 cents per share, depending on the broker, order size, and the level of service (algorithmic execution, research bundling, and so on). For a $50 stock, 3 cents per share represents 6 basis points per side, or 12 basis points round trip. On a portfolio with $100 million in annual trading volume, that is $120,000 in direct commission costs alone.

Options and Futures

While equity commissions have gone to zero at most retail brokers, options contracts still typically carry a per-contract fee. As of 2026, most retail brokers charge between $0.50 and $0.65 per options contract. For a 10-lot trade, that is $5 to $6.50 per side. For futures, commissions at retail brokers range from roughly $1.25 to $2.25 per contract per side, depending on the product and broker.

3. Component 2: The Bid-Ask Spread

The bid-ask spread is the difference between the highest price a buyer is willing to pay (the bid) and the lowest price a seller is willing to accept (the ask). When you buy at the ask and later sell at the bid, you pay the full spread as a round-trip cost. For a single crossing (one side of a trade), you pay roughly half the spread.

Spread magnitudes vary enormously across securities. For SPY, the SPDR S&P 500 ETF Trust — the most liquid equity security in the world — the typical quoted spread is about $0.01 on a price around $500, which translates to roughly 0.2 basis points per side. For large-cap stocks like Apple or Microsoft, quoted spreads are typically $0.01 to $0.02, translating to 0.5 to 1 basis point.

Move into small-cap territory and spreads widen dramatically. A stock trading at $10 with a $0.05 spread costs 50 basis points per round trip. For illiquid micro-caps with spreads of $0.10 to $0.50 on a $5 stock, the spread cost alone can be 200 to 1,000 basis points round trip. This is often the dominant cost for strategies that trade in the small-cap universe.

Effective vs. Quoted Spread

The quoted spread (also called the NBBO spread, for National Best Bid and Offer) is a snapshot. The effective spread is the actual cost incurred, measured as twice the distance between the trade price and the midpoint of the bid-ask at the time of execution. Effective spreads can be narrower than quoted spreads when trades receive price improvement (common with PFOF arrangements at retail brokers) or wider when large orders move the market.

4. Component 3: Market Impact

Market impact is the price movement caused by your own trading activity. When you place a buy order, you consume available liquidity on the ask side. If your order is larger than the quantity available at the best ask, you push the price up. The more you buy, the higher the price goes. This is the single largest cost for institutional traders and the primary reason why backtested strategy returns degrade as portfolio size grows.

The Foundational Model: Almgren & Chriss (2001)

The academic study of optimal execution was transformed by Robert Almgren and Neil Chriss in their 2001 paper “Optimal Execution of Portfolio Transactions” published in the Journal of Financial Markets, Vol. 3, No. 1. The paper formalized the tradeoff between two competing costs: the market impact of trading quickly (temporary and permanent price impact) and the timing risk of trading slowly (exposure to adverse price moves while your order is only partially filled).

Almgren and Chriss modeled the execution of a large order as a series of smaller child orders spread over a trading horizon. The key insight is that there is an efficient frontier of execution: aggressive schedules (trading fast) minimize timing risk but maximize impact cost, while passive schedules (trading slow) minimize impact but maximize timing risk. The optimal schedule depends on the trader’s risk aversion, the stock’s volatility, and the shape of the impact function.

The Square-Root Impact Model: Gatheral (2010)

Empirical studies of market impact consistently find that impact follows a concave function of order size. Jim Gatheral, in work presented around 2010 and subsequently published, proposed the square-root impact model, which has become a standard in the industry:

Impact ≈ σ × √(Q / V)

Where σ is the daily volatility of the stock, Q is the order quantity (in shares), and V is the average daily volume (in shares). The square root means that impact grows sublinearly with order size. If you double the order size, market impact increases by a factor of approximately √2 ≈ 1.41 — a 41% increase, not a 100% increase.

This has profound implications. It means that splitting a large order into smaller pieces and trading them over time is always cheaper than executing all at once (holding timing risk aside). It also means that market impact is disproportionately severe for illiquid stocks: a $1 million order in a stock with $5 million average daily volume (20% participation rate) has an impact of roughly σ × √0.20 ≈ 0.45σ, which for a stock with 2% daily volatility translates to about 90 basis points of price impact.

Why This Kills Strategies at Scale

Kenneth Korajczyk and Ronnie Sadka addressed this directly in their 2004 paper “Are Momentum Profits Robust to Trading Costs?” published in the Journal of Finance, Vol. 59, No. 3. They showed that the profitability of momentum strategies — buying recent winners and selling recent losers — is highly sensitive to fund size. For small portfolios, momentum generates attractive returns. But as assets under management grow, the market impact of entering and exiting positions in the less liquid stocks that drive momentum returns erodes profitability. At sufficiently large scale, the strategy becomes unprofitable after costs.

This finding generalizes beyond momentum. Any strategy that relies on trading less liquid securities or that requires high turnover faces a natural capacity limit imposed by market impact. This is why many successful hedge funds close to new investors well before their AUM approaches the theoretical capacity of their strategies.

5. Component 4: Slippage

Slippage is the difference between the price at which you expect to execute a trade and the price at which it actually fills. It encompasses several sub-effects: the price may move between when your system generates a signal and when the order reaches the exchange (latency slippage); the market may be moving fast and your limit order may need to be repriced (quote instability); or your market order may fill at a worse price than the last observed quote (stale data).

For algorithmic strategies, latency slippage is a function of infrastructure. A signal generated at time t based on a quote observed at time t − δ is already stale by δ milliseconds. In quiet markets, this barely matters. During high-volatility events — earnings releases, Fed announcements, geopolitical shocks — prices can move multiple ticks in milliseconds, and slippage can be catastrophic.

The practical implication is that backtests that use close prices or even minute-bar prices to determine fills are systematically optimistic. A more realistic approach is to assume execution at the open of the bar after the signal is generated, plus some additional slippage buffer. For daily strategies, assuming execution at the next day’s open with 5 to 10 basis points of slippage is a reasonable starting point for liquid names.

6. Component 5: Opportunity Cost

The most insidious cost is the one you never see in your execution reports. Opportunity cost arises when you use limit orders to avoid paying the spread and the order goes unfilled. The trade you did not take may have been a winner. This is sometimes called the “implementation shortfall” framework, originally proposed by Andre Perold in 1988.

Consider a strategy that generates a buy signal for a stock at $50.00. You place a limit order at $49.90 to save 20 basis points on the spread. The stock never dips to $49.90 and instead rallies to $55.00 over the next month. Your limit order saved you nothing because it was never filled, and you missed a 10% gain. The opportunity cost of that unfilled order is $5.00 per share — far more than the $0.10 you were trying to save.

Research consistently shows that the opportunity cost of unfilled orders dwarfs the execution cost savings for strategies with positive expected returns. This is why most institutional execution algorithms default to aggressive schedules (trading at or through the spread) rather than posting passive limit orders — the cost of missing the trade exceeds the cost of crossing the spread.

7. How to Model Transaction Costs in Backtests

Given the complexity of real-world transaction costs, how should a backtester model them? Here is a practical framework, ordered from simplest to most realistic:

Level 1: Fixed Basis Points

The simplest approach is to subtract a fixed cost per trade. Reasonable estimates:

• Liquid large caps (S&P 500 components): 10 basis points per side (20 bps round trip)
• Mid caps: 15–25 basis points per side
• Small caps: 25–50 basis points per side
• Micro caps / illiquid: 100+ basis points per side

These numbers should include spread, impact, and slippage combined. For a first-pass sanity check, this approach is adequate. If a strategy is not profitable with 20 bps round-trip costs on large caps, it is unlikely to work in production.

Level 2: Volume-Dependent Impact

A better approach uses the Gatheral square-root model to estimate impact as a function of order size relative to daily volume. This is critical for strategies that trade across a range of market capitalizations or that size positions based on signal strength (larger positions in higher-conviction names).

total_cost_per_side = spread_cost + sigma * sqrt(Q / V) + fixed_slippage

Where spread_cost is half the bid-ask spread (estimated or measured), sigma is daily volatility, Q is order size in shares, V is average daily volume, and fixed_slippage is a small buffer (2–5 bps) for latency and timing effects.

Level 3: Full Almgren-Chriss Simulation

For institutional backtests, the gold standard is to simulate the execution process itself. Break each order into child orders, model temporary and permanent impact separately, simulate the limit order book, and account for the time-varying nature of liquidity (lower volume at open and close, higher around the close auction). This level of modeling is typically only necessary for strategies managing hundreds of millions of dollars or more.

8. The Turnover Problem

Even modest per-trade costs compound dramatically with high turnover. Portfolio turnover is the fraction of the portfolio that is replaced over a period, typically annualized. A turnover of 100% means the entire portfolio is turned over once per year. A turnover of 500% means the portfolio is turned over five times per year — typical for a short-term momentum or mean-reversion strategy.

The arithmetic is straightforward:

Annual cost drag = Turnover × Cost per round trip

A strategy with 200% annual turnover at 20 basis points per round trip costs 400 basis points (4%) per year in transaction costs. If the gross alpha of the strategy is 8%, fully half of it is consumed by costs. At 500% turnover, the drag is 10% per year — enough to destroy almost any strategy.

This is why low-turnover strategies — such as insider-following strategies with holding periods of weeks to months — have a structural advantage. An insider-following strategy with 150% annual turnover at 20 bps round trip costs only 3% per year, leaving much more room for net alpha.

9. Post-Publication Decay and Transaction Costs

David McLean and Jeffrey Pontiff examined this phenomenon in their 2016 paper “Does Academic Research Destroy Stock Return Predictability?” published in the Journal of Finance, Vol. 71, No. 1. They studied 97 return-predicting anomalies documented in academic journals and found that portfolio returns for these anomalies decline by approximately 32% out of sample (after the sample period ends) and by a further 26% post-publication (after the paper is published).

McLean and Pontiff argue that this post-publication decay has two causes. First, arbitrageurs learn about the anomaly from the published paper and begin trading it, which reduces returns. Second, and critically, transaction costs explain a significant portion of the apparent anomaly returns in the first place. Many anomalies concentrate in small, illiquid stocks where trading costs are highest. When realistic costs are applied, the profitability of these anomalies is substantially reduced — and in many cases eliminated entirely.

10. Practical Implications for Strategy Design

The existence of transaction costs imposes a set of design constraints on any strategy that aims to be profitable in production:

• Target liquid names. Strategies focused on S&P 500 or Russell 1000 components face far lower execution costs than those trading micro-caps. The spread and impact alone can be 10x to 100x higher in illiquid names.
• Minimize turnover. Every trade costs money. Strategies that hold positions for weeks or months inherently incur less cost than those that rebalance daily. Use turnover budgets to constrain optimization.
• Size positions relative to liquidity. A $50,000 position in a stock that trades $500,000 per day is a 10% participation rate — expect meaningful market impact. The same $50,000 in a stock trading $50 million per day is 0.1% participation — impact is negligible.
• Use implementation shortfall benchmarks. When executing live, measure execution quality against the arrival price (mid-quote at the time the decision was made). This captures all costs: spread, impact, slippage, and timing.
• Stress-test with conservative assumptions. If a strategy is marginally profitable at 20 bps per side, it will almost certainly lose money live. Demand a healthy margin of safety between gross alpha and estimated costs.

11. Bringing It All Together

Transaction costs are not an afterthought — they are a first-order consideration in strategy design. The five components (commissions, spread, market impact, slippage, and opportunity cost) interact in complex ways, and their magnitudes vary by orders of magnitude across different market segments. A strategy that looks brilliant in a zero-cost backtest may be worthless after costs.

The good news is that understanding these costs gives you an edge over the many practitioners who ignore them. By targeting liquid securities, minimizing turnover, modeling impact realistically, and maintaining a healthy cushion between gross alpha and estimated costs, you can build strategies that survive the transition from backtest to production.

The best quant strategies are not the ones with the highest gross returns in backtests. They are the ones that generate the most alpha after every dollar of transaction cost has been paid.