Originally Published: · Last Updated:
Neither SMA nor EMA is the automatic winner in a crossover strategy. EMA reacts more strongly to recent prices, but that does not guarantee more trades, fewer false signals, or better net returns. In a dated SPY illustration using six fast/slow window pairs, EMA produced more position changes in three pairs, fewer in two, and the same number in one.
Judge the exact rule by whether it survives next-open execution, trading costs, taxes, and a different market period. Barber and Odean’s household study supports caution about heavy trading, but it did not compare SMA with EMA and cannot be converted into a fixed annual penalty for every crossover strategy.
A trader looking at the same chart can make two reasonable choices. Use an SMA to smooth the noise, or use an EMA to react sooner. A quicker signal answers only the timing question. It says nothing yet about whether the trade helps.
A faster signal can exit a real decline earlier. It can also move to cash during a brief dip and buy back after the rebound. A slower signal can avoid some of those reversals, but it can stay exposed longer when a trend truly changes. The result depends on the market path, the selected windows, the execution rule, and the cost of changing positions.
SMA vs EMA Crossover: The Comparison in One Table
An SMA gives every observation inside its window the same weight. An EMA gives more weight to recent observations and lets older observations fade gradually. That difference changes the shape and timing of the line. Performance still comes from the full rule.
SMA vs EMA Crossover Decision Table
| Question | SMA | EMA | What actually decides |
|---|---|---|---|
| How are prices weighted? | Equal weight inside the window | More weight on recent prices | The chosen window and smoothing convention |
| Which reacts sooner? | Usually slower for the same nominal window | Usually faster for the same nominal window | How large and persistent the price move is |
| Which creates more trades? | No fixed winner | The fast/slow pair, market path, and crossover definition | |
| Which has fewer whipsaws? | No fixed winner | Trend persistence and the distance between the two lines | |
| Which earns more after costs? | Must be tested | Execution lag, costs, taxes, cash return, and sample period | |
Plain-language distinction: an EMA is more responsive, not more informed. It sees the same historical prices as an SMA and simply assigns them different weights.
That distinction matters because many charting tutorials treat responsiveness as if it were an edge. It is only an edge when the earlier reaction is useful more often than it is costly.
This analysis applies to: long-or-cash crossover rules on liquid US equity funds or indexes, where a faster average crossing a slower average changes the portfolio position.
It does not settle: intraday execution, leveraged products, short-selling systems, multi-asset trend following, or institutional strategies with different cost and risk controls.
Why Faster Does Not Mean Better
EMA reduces one kind of delay. It can recognize a fresh move sooner because recent observations carry more weight. That advantage is real, but it comes with a tradeoff: a short-lived price move also receives more influence.
Imagine a fast average crossing below a slow average during a sharp three-day decline. An EMA pair may move to cash first. That helps if the decline continues. It hurts if prices reverse before the slower SMA pair ever changes position.
The same tradeoff appears at re-entry. EMA may buy back sooner after a recovery, but it may also complete a sell-buy round trip that the SMA never made. The result cannot be inferred from line speed alone.
The research record also changes with the model and test design. Zakamulin’s 2018 reanalysis found that a previously reported moving-average advantage was largely created by look-ahead bias, and the corrected strategy was statistically indistinguishable from its corresponding buy-and-hold benchmark. That paper addressed a timing error in a specific study; it did not prove that every SMA and EMA rule has the same result.
Other models reach conditional conclusions. A 2022 Markov-switching paper by Zakamulin and Giner derives EMA as optimal under its two-state assumptions, while their 2023 transaction-cost model finds that an optimal practical rule can resemble an SMA crossover. For a retail investor, the preferred average therefore depends on the assumed market process and cost structure.
Model sources: Zakamulin’s paper archive · 2022 EMA model · 2023 transaction-cost model
For a retail investor, these papers do not settle the question by themselves. A moving-average rule may serve as a risk-management framework, but it still has to be judged against the job assigned to it.
What Barber and Odean Actually Measured
Barber and Odean’s 2000 study is strong evidence that heavy household trading was associated with poor net performance in its sample. The paper studied household portfolios, so its regression estimates cannot serve as an SMA-or-EMA fee schedule.
The study analyzed common-stock portfolios held by 66,465 households at one large discount brokerage from 1991 through 1996. The researchers sorted households into five groups based on mean monthly turnover. Quintile 1 contained the lowest-turnover households, and quintile 5 contained the highest.
Selected Barber and Odean Turnover Quintiles
| Household group | Mean monthly turnover | Net Fama-French intercept per month | Correct interpretation |
|---|---|---|---|
| Q1, lowest turnover | 0.19% | −0.061% | Risk-adjusted net household result in the sample |
| Q4 | 5.98% | −0.464% | Fourth turnover quintile, not the highest group |
| Q5, highest turnover | 21.49% | −0.864% | Highest-turnover household group |
The Q4 Fama-French intercept of −0.464% per month is a risk-adjusted performance estimate for that household group. It can reflect historical commissions and spreads, security selection, trade timing, portfolio characteristics not fully captured by the model, and investor behavior. Subtracting it from a hypothetical 7% return would treat a regression outcome as though it were a fee quote.
The original paper separately estimated trading costs. In the early-1990s brokerage sample, the authors reported substantial commissions and bid-ask spreads. Those figures help explain why frequent trading was expensive then, but the brokerage regime has changed too much to transplant them unchanged into a modern ETF projection.
The study also did not record whether a household used an EMA crossover, an SMA crossover, a newsletter, discretionary stock picking, or no formal signal at all. It therefore cannot establish that both crossover types lose the same amount.
Primary source: Barber and Odean, Trading Is Hazardous to Your Wealth, especially Table V and the transaction-cost discussion.
Measure turnover from the strategy itself instead of borrowing one household-quintile alpha as its exact cost.
A Reproducible SMA vs EMA Backtest
To test the direct comparison, TheFinSense ran a fixed six-pair illustration on a pinned third-party SPY OHLC series from January 2010 through December 2019. The six pairs were fixed before the run; no winner was selected after seeing the results. The purpose was to see whether EMA necessarily creates more crossovers or the same wealth outcome.
Rule: hold SPY when the fast moving average is above the slow moving average; otherwise hold cash.
Execution: calculate the signal from the file-provided close on day t and apply any position change at the next available open on day t+1.
Costs: subtract 25 basis points for each position change. Cash earns 0%. Taxes and additional slippage are excluded.
Parameters: six fast/slow pairs chosen before running the comparison: 5/20, 10/30, 20/50, 50/100, 50/200, and 100/200.
Reproducibility: the third-party OHLC file is pinned to a specific Git commit. The script records its SHA-256 hash, the execution convention, and every output row.
Dated SPY Illustration: SMA vs EMA Position Changes and Net CAGR
| Fast/slow pair | SMA changes | EMA changes | SMA net CAGR | EMA net CAGR | Buy-and-hold CAGR |
|---|---|---|---|---|---|
| 5/20 | 145 | 153 | −0.17% | −0.26% | 13.84% |
| 10/30 | 81 | 95 | 3.30% | 1.71% | 13.79% |
| 20/50 | 53 | 35 | 5.54% | 6.88% | 13.24% |
| 50/100 | 19 | 11 | 8.79% | 9.76% | 14.28% |
| 50/200 | 9 | 9 | 9.23% | 9.44% | 13.93% |
| 100/200 | 7 | 9 | 9.16% | 8.94% | 13.93% |
Trade counts split across the six pairs. EMA produced more position changes in the 5/20, 10/30, and 100/200 pairs. SMA produced more in the 20/50 and 50/100 pairs. The 50/200 pair tied.
The return ranking also moved. EMA finished ahead for the 20/50, 50/100, and 50/200 pairs. SMA finished ahead for the 5/20, 10/30, and 100/200 pairs. Buy-and-hold led all six in this particular period. None of those observations proves what will happen in the next decade. They show why the answer cannot be determined from the moving-average label alone.
A researcher can run hundreds of window pairs and publish the winner. That procedure turns noise into a polished equity curve. Fix the parameter family first, use an execution lag, charge realistic costs, and reserve a separate period for evaluation.
For a broader checklist, see TheFinSense’s guide to technical analysis backtest data. The golden cross win-rate analysis shows why even a familiar 50/200 rule needs a defined holding period and benchmark.
Backtest data: Pinned third-party SPY CSV. Dataset SHA-256: f1682f176f9db69a
The source is a third-party price-data convenience file rather than a licensed CRSP series or an issuer-certified total-return record. The test uses one ETF proxy and one decade, assumes zero return on cash, excludes taxes and extra slippage, and does not correct for a wider parameter search. It can refute a universal SMA-versus-EMA claim; it cannot forecast live performance.
A live rule needs more data, a holdout period, and implementation testing beyond this table.
How to Choose a Crossover Rule
A crossover rule should be chosen as a complete decision system. The average type is only one input.
1. Define the job before the indicator
Decide whether the rule is meant to reduce drawdowns, preserve participation in long trends, limit emotional decisions, or generate short-term trades. A rule built for drawdown control may accept long periods of underperformance. A short-term trading rule cannot be judged by the same standard.
2. Specify every execution detail
Write down the price field, fast and slow windows, signal timing, entry price, exit price, cash return, transaction cost, tax treatment, and whether short positions are allowed. A backtest without these details is not reproducible.
3. Measure the two costs separately
Count direct turnover costs such as spread, slippage, and taxes. Then measure opportunity cost from sitting in cash during rebounds. Combining both into a single borrowed “drag” number hides which part of the strategy failed.
4. Check whether nearby parameters behave similarly
A rule is more credible when nearby parameter pairs produce similar behavior. If 19/49 looks excellent and 20/50 collapses, the apparent edge may be parameter luck rather than a durable signal.
5. Use a holdout period
Select or design the rule on one period, then evaluate it on data that was not used for selection. Zakamulin’s look-ahead-bias correction is a reminder that clean-looking historical performance can be created by information the trader could not have known at the time.
6. Compare the rule with a realistic alternative
Buy-and-hold is one benchmark, but it may not match the investor’s objective. A drawdown-control rule can be compared on return, maximum drawdown, time in cash, tax events, and the probability that the investor could follow it. The benchmark should reflect the decision the rule is meant to improve.
Investors using charting platforms should also confirm that the displayed indicator settings match the backtest. TheFinSense’s TradingView settings guide covers adjustment choices that can quietly change a signal.
Frequently Asked Questions
Is EMA always faster than SMA?
For the same nominal window, EMA generally reacts more strongly to recent prices. That does not mean it always changes a crossover position sooner, because the result also depends on the second average and the path of prices.
Does EMA always create more crossover trades?
No. In the six-pair SPY illustration in this article, EMA created more position changes in three pairs, fewer in two, and the same number in one. Trade count is a property of the complete rule, not the weighting label alone.
Did Barber and Odean compare SMA with EMA?
No. Their study compared household portfolios grouped by turnover. It did not identify the households’ trading signals, so it cannot establish equal losses for SMA and EMA crossover users.
Can the Q4 intercept be used as an annual transaction-cost rate?
No. The −0.464% monthly Q4 figure is a net Fama-French regression intercept for a historical household group. It is not a pure fee estimate and should not be subtracted mechanically from a projected portfolio return.
Which moving average should a beginner use?
Start with the rule you can define and test clearly. An SMA is easier to explain, while an EMA is more responsive. Neither should be used with real money until the exact windows, lag, costs, and risk behavior have been tested.
SMA vs EMA Crossover: Bottom Line
EMA reacts faster because it weights recent prices more heavily. SMA smooths the same history with equal weights. That is the technical difference. The investment result depends on the rest of the rule.
Barber and Odean provide a strong warning about heavy household trading in a high-cost historical brokerage environment. The study does not assign a fixed annual penalty to an EMA or SMA crossover. A direct test is needed, and even that test applies only to its stated market, period, parameters, and assumptions.
Before choosing a line, write the full rule and ask one question: does the apparent benefit remain after next-open execution, realistic costs, nearby parameter tests, and a holdout period? If you cannot show the answer, do not trade the rule yet.
YOUR TURN
For the crossover rule you are testing, how many position changes remain after you move execution to the next available open and add your actual trading costs?
Editorial transparency: AI-assisted tools supported source comparison, code execution, and drafting. The publication decision and responsibility for the final article remain with Danny Hwang.
Disclosure: No broker, fund issuer, or charting platform paid for this analysis. Examples are educational and are not personalized investment advice.
Update history
-
v2.0
2026-07-22
CORRECTIONReplaced the unsupported universal turnover-drag projection with a measurement-correct reading of Barber and Odean and a reproducible SMA-versus-EMA comparison.
-
v1.0
2026-05-08
PUBLISHOriginal publication.
Educational quantitative analysis based on published data. Not investment, tax, or legal advice. Consult a licensed professional before acting on any calculation. About TheFinSense.