Technical Analysis.

No tool is more popular or more divisive than technical analysis (TA). To some, it's a map for reading market psychology; to others, it's little better than reading tea leaves. Old-guard academics dismissed it; practitioners swear by it. The truth sits somewhere between the two — and, rarely stated plainly, it depends heavily on which technical analysis you mean, and how it was tested.

This article summarizes what the empirical evidence actually shows: which parts of TA survive rigorous testing, which parts collapse, and why — with specific notes on the IDX market.

The big picture: TA isn't one thing, it's a spectrum

The biggest mistake in this debate is treating "technical analysis" as a single package. In reality it spans a wide spectrum:

• Systematic — rules that can be written down and tested: momentum, moving averages, trend-following. It follows numbers, not interpretation.
• Discretionary — pattern reading and "feel": head-and-shoulders, hand-drawn trendlines, "support" identified after the fact. It depends on the interpreter.

The empirical evidence differs sharply across this spectrum. The systematic end has reasonably solid support; the discretionary end mostly doesn't. So the right question isn't "is TA useful?" but "which TA, and can it be falsified?"

Starting point: what theory says

The Efficient Market Hypothesis (Fama, 1970) — in its most basic form, technically called the weak form — holds that stock prices already absorb the entire history of past prices. The consequence is that studying past price patterns (the core of technical analysis) should, in theory, offer no edge at all. That's a hard theoretical position.

But real markets aren't a perfect idealization. Measurable anomalies do exist and persist. So theory says "it shouldn't work," while the evidence says "it depends what you mean." Both can be true at once — and that's exactly where the nuance lies.

What the evidence supports: momentum and trend

The most academically solid part of TA isn't chart patterns at all — it's momentum.

• Cross-sectional momentum. Jegadeesh & Titman (1993) showed that stocks that outperformed over the past 3–12 months tend to keep outperforming over the following months, and vice versa. It's one of the strongest and most frequently replicated anomalies across markets and decades.
• Time-series momentum (trend). Moskowitz, Ooi & Pedersen (2012) found that the tendency of prices to continue in their direction holds consistently across many asset classes — the empirical foundation behind trend-following strategies.

A note that's often glossed over: momentum has a dark side. It can collapse sharply when the market reverses — the "momentum crash" phenomenon (Daniel & Moskowitz, 2016) — so its attractive average returns come with real tail risk. This is a statistical edge, not a money machine.

Where the evidence is weak or mixed: classic rules and patterns

This is where systems sold with promises of certainty start to crack.

• Moving-average and breakout rules. Brock, Lakonishok & LeBaron (1992) once found predictive power in the Dow index (1897–1986). But that finding didn't hold up under statistical correction: Sullivan, Timmermann & White (1999) showed that once you account for data snooping — the fact that the "best" rule was cherry-picked from many tried — the edge shrinks and becomes fragile out of sample. Bajgrowicz & Scaillet (2012) re-tested thousands of rules on the DJIA from 1897–2011 using a False Discovery Rate method and reached two damning conclusions: investors could never have picked the future-winning rule in advance, and whatever edge existed disappears once even small transaction costs are included.
• Visual patterns. Lo, Mamaysky & Wang (2000) tested algorithmic recognition of classic patterns (like head-and-shoulders) and found that some patterns are indeed statistically distinguishable from pure chance — they carry "extra information." But their practical economic value is limited and disputed; the gap between "statistically different from random" and "usable for profit after costs" is very wide.

The most complete picture comes from Park & Irwin (2007), who surveyed 95 modern studies: 56 found positive results, 20 negative, 19 mixed. Early studies found profits in currency and futures markets — but not in equities — and much of that edge faded starting in the early 1990s. The takeaway: there is a signal there, but it is fragile, inconsistent, and shrinking over time.

Why TA "works" in stories but fails under testing

The gap between testimonials and data almost always comes from the same handful of causes — and they line up neatly with the backtesting traps:

• Data snooping. Try 200 indicators, then show off the one that happened to fit. Out of 200 random attempts, something is bound to look brilliant.
• Unfalsifiable. Patterns are identified after the fact. There's always a trendline you can draw to fit the past — and that itself is a red flag, not evidence.
• Ignoring costs. Frequent signals mean frequent trades; fees, sales tax, and slippage grind thin edges down to nothing.
• Non-stationarity and reflexivity. The more people trading on the same signal, the thinner its edge becomes. This is the core of the Adaptive Markets Hypothesis (Lo, 2004): a strategy's edge evolves and decays as market participants' behavior changes.

IDX context: why discretionary TA is even more fragile here

Several features of the Indonesia Stock Exchange make discretionary pattern-reading riskier than the global narrative suggests:

• Thin liquidity in second- and third-tier stocks. Patterns can form "by coincidence" on small volume, and your own large order can move the price — corrupting the very signal you thought was objective.
• ARA/ARB (auto rejection). IDX enforces daily upper and lower price limits. A "breakout" signal can appear right as a stock is locked at its limit — in reality, you won't get filled at the price your chart assumes.
• Wide spreads and slippage on illiquid stocks erode small edges even faster.
• The "bandarmology" and "pump group" ecosystem. Pattern narratives in social-media groups are often a pump-and-dump tool rather than analysis — here, charts are used to justify a story, not to test one.

How to use TA without fooling yourself

The good news is that the evidence-based part of TA can be used — as long as it's subjected to the same discipline as testing any other strategy:

• Favor systematic over discretionary. A rule that can be written down can be tested; a "feel" cannot.
• Define the rule before seeing results, then test out of sample. If it collapses on data it hasn't seen before, that's memorization, not edge (see our Backtesting article).
• Include real IDX costs — fees, tax, slippage — from the start.
• Treat it as a probabilistic edge, not a prophecy. Think in P10/P50/P90 ranges, not one guaranteed line.
• Be suspicious of anyone who only shows winning examples. A properly tested system shows its failures too.

Closing

Technical analysis is not baseless fortune-telling: part of it — particularly momentum and trend — survives even the strictest tests. But it isn't magic either: most discretionary pattern-reading fails to clear that same evidentiary bar, and nearly every system that promises certainty collapses once it's tested honestly, charged real costs, and corrected for luck.

The dividing line is always the same, for TA or any other tool: can it be falsified and tested, or not. That's the standard we hold at Sobat Investor — weigh every tool against the same evidentiary bar, then state its uncertainty as it is.

References & Further Reading

This article is educational and does not constitute investment advice. Investment decisions and their risks are entirely your own responsibility. Past performance does not guarantee future results.