What the evidence says about technical trading

The chapter this subject exists for. Eight chapters described the methods; this one reports what happened when they were tested.

The state of the literature, stated fairly

The honest summary is not "it doesn't work". It is more interesting than that, and the source states it directly:

Numerous empirical studies have investigated the profitability of technical trading rules in a wide variety of markets, and many of them found positive profits. Despite positive evidence about profitability and improvements in testing procedures, skepticism about technical trading profits remains widespread among academics mainly due to data snooping problems.

Both halves matter. Positive results exist and are numerous. And the scepticism is not a refusal to look — it is a specific objection to what the positive results can establish.

Why positive results are discounted

Chapter 8 introduced the problem; here is the definition in full:

Data snooping occurs when a given set of data is used more than once for purposes of inference or model selection. If such data snooping occurs, any successful results may be spurious because they could be obtained just by chance with exaggerated significance levels.

And the ways it enters, in the same source:

The blatant form — an ex post, in-sample search for profitable rules.

The subtle forms — a dataset repeatedly used to search over "profitable families of trading systems, markets, in-sample estimation periods, out-of-sample periods, and trading model assumptions including performance criteria and transaction costs."

The collective form — researchers each choosing one method on the same dataset are "collectively snooping the data", and this is "potentially even more dangerous because it is not easily recognized by each individual researcher."

Read that list against chapters 2 through 6. Every parameter named there — the interval, the window length, the lookback, the thresholds — is a dial in that search space. A literature testing thousands of rule-and-parameter combinations on the same price histories will produce winners whether or not any effect exists. That is arithmetic, not cynicism.

The test that addresses it

The remedy the economics literature proposes is simple and demanding: replicate the earlier result on new data. A rule found by searching one period, then applied unchanged to a period that did not exist when the rule was chosen, cannot have been fitted to the second period.

Park and Irwin do exactly this — confirming a prior study's results, then applying the original testing procedure to fresh data. And the finding:

Results indicate that for various futures contracts and technical trading systems tested, technical trading profits have gradually declined over time. In general, substantial technical trading profits in the early 1980s are no longer available in the subsequent period.

That is the sentence to remember from this subject. Not that the rules never worked — the early profits appear to have been real. That they stopped working, and the decay shows up precisely when you test on data the rules were not chosen from.

Why an effect would decay

The decay is not mysterious, and it is consistent with how markets are supposed to behave.

A published profitable rule attracts users. If a rule works and becomes known, participants act on it, and acting on it moves prices toward where the rule said they would go — which removes the opportunity. The Herding, narratives and bubbles chapter describes the same self-defeating mechanism.

Costs fell and competition rose. Faster execution and cheaper data let more participants compete away smaller edges.

Markets changed structurally. The futures markets of the early 1980s had different participants, liquidity and technology.

So the pattern "it worked, then it didn't" is what an efficient-enough market looks like from the inside — which is a more informative finding than either "it works" or "it never did".

The Indian evidence

The research above concerns rules tested on historical data. The Behavioural finance subject supplies the other half: what happens to the people actually trading.

SEBI's study of the equity derivatives segment — where most retail technical trading in India occurs — found 91% of individual traders incurred net losses in FY25, with the loss-making share stable between 90% and 92% across FY22 to FY25 even as participation grew toward 96 lakh traders.

The stability of that share is the finding, as that chapter argued: if experience or technique improved outcomes, the proportion would fall as the cohort aged. It does not move.

And the mechanism is cost. Barber and Odean's survey found individual investors underperform "largely because of trading costs", with the most active quintile earning 11.4% a year against 18.5% for the least active. Technical methods, whatever their predictive content, are activity-generating by construction — they produce reasons to transact. That interacts badly with the single most robust finding about individual investors.

Working the problem

A study reports a rule earning 12% a year. What must you know?

On how the rule was chosen:

1. Was it specified before seeing the data, or found by searching it? If searched, over how many rules, parameters, markets and periods? The number of combinations tried determines how impressive 12% is.
2. Was it tested out-of-sample, on data not used in the search?
3. Has it been replicated on data generated since publication?

On what the 12% is measured against: 4. What did buy-and-hold return over the same period in the same instruments? A rule earning 12% while the market returned 15% has a negative result reported as a positive one. 5. What risk was taken? The source's standard is explicitly risk-adjusted: a rule is profitable only if its risk-adjusted profits exceed costs. 12% with twice the volatility is not 12%. 6. What was the worst drawdown, and would anyone have held through it? Risk-adjusted return in Measuring your return makes the point that a return you would not have survived is not available to you.

On implementation: 7. Transaction costs — the literature's bar, and the one that kills most results. Spread, brokerage, taxes, and in India, STT. 8. Turnover. How many trades a year? High turnover multiplies every cost. 9. Capacity and slippage. Did the test assume fills at the closing price? Real orders move thin markets. 10. Survivorship. Did the instrument universe include the ones that delisted?

The single most important of these is 1. Everything else can be checked from the paper; whether a search preceded the result often cannot. And a result from an unreported search is indistinguishable from a real effect by any amount of further analysis of the same data — which is why replication on new data, rather than cleverer statistics, is the remedy the literature settled on.

The point

Many studies found technical trading profits, and academic scepticism rests not on ideology but on data snooping — a dataset reused for inference or model selection yields successes obtainable by chance, and the search extends across rule families, markets, estimation periods and cost assumptions, often collectively and invisibly. The remedy is replication on new data, and when Park and Irwin applied it, technical trading profits had gradually declined, with substantial early-1980s profits no longer available later. Decay is what a competitive market should produce once a rule becomes known. Alongside that, 91% of Indian individual derivatives traders lost money in FY25 with the share stable for four years, and individual underperformance generally traces to trading costs — which methods that generate reasons to trade will tend to incur.