BLL 1992: Moving Average Rules, Data Snooping, and Transaction Costs

What Brock, Lakonishok & LeBaron Actually Found

In 1992, William Brock, Josef Lakonishok, and Blake LeBaron published a study that became one of the most-cited papers in technical analysis. They tested simple moving average crossover rules and trading-range break rules — a systematic form of support and resistance — on daily Dow Jones Industrial Average data spanning nearly a century. Their finding: buy signals tended to precede above-average returns and sell signals tended to precede below-average ones. For traders who had long been dismissed as chartists, this was a landmark.

The authors were careful about scope. They tested a specific set of rules, on one index, over one historical period. They were not claiming a universal law — they were documenting a pattern. That nuance got lost in the retelling, and it is the source of most of the confusion that followed.

The Data-Snooping Problem

Data snooping bias — also called data mining or rule-selection bias — occurs when enough rules are tested on the same dataset that some appear significant purely by chance. If you try hundreds of moving average parameter combinations on a single price series, a handful will produce impressive results even if the underlying process is random. The more rules you test, the higher the bar that any single result needs to clear before you can call it meaningful.

Subsequent researchers applied methods designed to correct for this, accounting for the full universe of possible technical rules rather than just the best-performing handful. When that correction was made, the excess returns from the strongest BLL-style rules shrank considerably. This is not a claim that moving averages are useless — it is a reminder that a result that looks good on one dataset, with pre-selected parameters, is very difficult to distinguish from a lucky draw.

The practical implication: a backtest on a single asset over a single historical period, using a small set of parameter choices, tells you almost nothing on its own. The honest test is whether an indicator generates consistent edge across a large, diverse set of assets and genuinely out-of-sample time windows.

Transaction Costs: The Detail That Changes Everything

BLL used conservative transaction cost assumptions, which was appropriate for the academic context. In practice, transaction costs — spreads, slippage, and commissions — are not zero. For strategies that generate frequent signals, they accumulate quickly. A moving average crossover that trades many times per year can look profitable before costs and unprofitable after them, particularly in less liquid markets.

This effect is not uniform. In highly liquid large-cap markets, transaction costs are genuinely low and their impact is modest. In smaller-cap stocks, thinly traded crypto tokens, or exotic currency pairs, wide spreads can absorb most of a strategy's gross edge. The only honest evaluation bakes in realistic costs from the start rather than optimizing on gross returns and adjusting afterward.

All backtests on IndicatorEdge include realistic transaction costs by design. A result that only works before costs does not appear in the rankings.

What 660,005 Out-of-Sample Backtests Show

IndicatorEdge ran 660,005 out-of-sample backtests across 903 assets and 382 indicators, covering four timeframes: 1-Hour, 4-Hour, Daily, and Weekly. The question for each test was simple: after realistic transaction costs, does this indicator beat buy-and-hold for this asset on this timeframe?

The headline result is that only 26% of all indicator-asset-timeframe combinations beat buy-and-hold. Across assets, 63% have at least one indicator that clears the bar — but knowing which one in advance, before the out-of-sample period, is the hard part. The median best Sharpe ratio across assets is 0.62, which represents a real but modest edge.

Moving averages appear across asset classes but do not dominate universally. MA Envelope leads the crypto category. EMA 100 Trend and T3 200 Trend appear among the top Index ETF indicators. But no single moving average rule tops every asset class — for Forex, the Fisher Transform leads with 17 top-asset wins; for Stocks, Fibonacci Pivots, Projection Bands, and Camarilla Pivots lead. The BLL-style claim of a broadly effective moving average rule does not survive a cross-sectional test at this scale. Browse assets to see what actually tops each one.

The Win-Rate Trap and Why It Matters

Some of the most popular indicators show a pattern that looks compelling until you examine it closely: high win rates paired with very low rates of actually beating buy-and-hold after costs. RSI Mean-Reversion posts a median win rate of 71.7% — it picks profitable trades more often than not across many assets. Yet only 10% of assets actually beat buy-and-hold with it. CCI shows a 71.0% median win rate with 9% of assets beating buy-and-hold. Money Flow Index is 72.2% and 9%. Holy Grail Confluence is 73.3% and 8%.

SMC-based approaches follow the same pattern. SMC: Liquidity Sweep shows a median win rate of 71.2% but only 8% of assets beat buy-and-hold with it — and no Smart Money Concepts indicator topped the rankings for any asset in our tests.

The disconnect between win rate and real edge over a passive hold is one of the oldest traps in technical analysis. An indicator can win most of its individual trades and still underperform buy-and-hold if winning trades are small, losing trades are large, or transaction costs erode the margin. The BLL critique applies in modern form: the question is never whether an indicator has posted an impressive number somewhere — it is whether it does so reliably, after costs, across a broad population of assets. Browse the full indicator rankings to see which ones actually clear that bar.