What Technical Indicators Cannot Tell You

We display technical indicators on every stock page. We also think most people misread them badly. Both things are true, and this piece is the second one.

The one fact that reframes everything

Every technical indicator is a function of past price. That is not a criticism — it is a definition. RSI, MACD, ADX and moving averages take price history in and put a transformed number out.

The consequence follows immediately: an indicator cannot contain information the price history does not already contain. It is a re-presentation. Useful, because humans read charts poorly and summaries help. But no indicator knows a company won a contract, lost a lawsuit, or is about to report bad numbers.

So when a stock gaps 15% on news, no indicator "missed" it. They cannot see news. They only saw the aftermath.

"Overbought" does not mean sell

The most expensive misunderstanding in retail investing.

RSI above 70 gets called overbought. The word implies "too high, due a fall." It means nothing of the sort. It means recent gains have been large relative to recent losses. That is all it means.

In a strong uptrend RSI can stay above 70 for weeks. Stocks in the middle of major re-ratings are overbought the entire way up. Selling because a line crossed 70 has taken people out of their best positions early, repeatedly, for decades.

The mirror error is worse: buying because RSI dropped below 30. A collapsing company is oversold all the way to zero. "Oversold" is not a floor.

Indicators lag, by construction

MACD is built from moving averages. Moving averages are backward-looking by definition — averages of what already happened. So MACD tells you a trend changed after it changed.

Sometimes that is fine; confirmation has value. But do not confuse a lagging confirmation with a prediction. By the time the histogram flips, the move is underway and part of it is gone.

The parameters are chosen, not discovered

RSI-14. MACD 12/26/9. The 50-day and the 200-day moving average. These get quoted so often they start to feel like constants of nature. They are not. They are conventions, most of them fixed decades ago by people plotting charts by hand, when arithmetic was expensive.

The tempting fix is to optimise: run every combination of lookbacks over ten years of KSE-100 history and keep the best pair. You will find one. You will also have fitted the noise of that decade, and the next decade has different noise.

There is one honest defence of the defaults, and it is far weaker than it sounds. A great many people watch 14, 50 and 200, and orders cluster where attention clusters. A break of the 200-day gets written up, which draws buyers or sellers to the level, which can make the level matter for a session. That is a real effect — but notice the claim. The number matters because people watch it, so it stops mattering the moment they stop, and it never applied to the stock nobody watches. On the PSX, where much of daily turnover sits in a few dozen names, that caveat removes most of the market.

They fail hardest in sideways markets

Trend indicators need a trend. Without one they whipsaw — signal, reverse, signal, reverse — and each false signal costs commission and tax.

This matters especially on the PSX, which spends long stretches range-bound on thin volume. A strategy backtested through a trending stretch can look excellent and then bleed for a year when the market goes quiet.

ADX is the honest counterweight: below roughly 20, treat trend signals as noise. The most valuable output of a technical toolkit is often "there is nothing here."

Thin volume breaks them

Indicators assume the price series means something — that it reflects many participants continuously repricing.

For a PSX small cap trading a few thousand shares a day, that assumption fails. A handful of trades sets the close. RSI computed on that series is arithmetically valid and informationally empty. It measures the behaviour of three people.

Before reading any indicator, look at volume. If it is thin, the indicator is describing noise with a decimal point. Average daily volume sits on every stock page here; our guide to reading a stock page covers how much weight each number deserves.

Circuit breakers distort the series

PSX price caps mean the close is sometimes not a real clearing price — it is the boundary. A stock locked at upper circuit did not "close at" that price in the sense the indicator assumes; there were buyers with nothing available.

Feed a circuit-locked series into any indicator and the output is a number computed from a constraint, not from supply and demand. Every indicator on this site inherits that flaw, including our model's inputs.

What indicators cannot see on the PSX specifically

Two blind spots deserve naming, because they are local and large.

Corporate actions. Bonus issues, rights issues and splits change the share count without changing the business. A 20% bonus issue mechanically lowers the quoted price the morning it goes ex — the same ownership is simply divided into more pieces. An unadjusted series records that as a fall, and every indicator agrees: the average rolls over, RSI sinks, MACD crosses down. Nothing happened. Serious providers adjust for these events and ours does, but verify rather than assume with any source. A sharp drop with no news behind it is worth checking against the company's PSX announcements first.

The macro layer. Pakistan's market is repriced from the top down more often than most. The policy rate peaked at 22% in 2023-24, and its path since has moved bank margins and the appeal of equities against a term deposit — with no chart pattern giving notice. An IMF review passing or stalling, a rupee move, a tariff decision: each has repriced whole sectors in a session, and the indicator only learned of it the next day. The macro cycle piece covers what really drives the index over years, and the banking sector explainer shows the clearest transmission, since bank earnings track the policy rate almost mechanically.

The chart-pattern problem

Head and shoulders. Flags. Ascending triangles. Cup and handle. These are older than the oscillators and, for many investors, more persuasive — you can see them, which feels like evidence.

The difficulty is that humans are pattern-completion machines. We find a face in two dots and a line. Generate a chart from random numbers, hand it to a confident chart reader, and a textbook head and shoulders comes back, because the eye supplies the shape the data only hinted at. That is not a personal failing; it is how vision works, and it does not switch off because the numbers are real prices.

Give the same chart to two experienced people and you will regularly get opposite readings. The falling wedge one of them calls an exhausted downtrend is, to the other, a plain continuation. Neither can settle it, because nothing was measured: the trendlines were drawn by hand, and the pattern was named after it completed.

Contrast the one thing on a chart that is genuinely countable: volume confirmation. A breakout on four times average volume and a breakout on half of it are different events, and the difference is a fact rather than an interpretation. Volume gives you no direction and is not a signal on its own. But it counts participation, which is the only part of the picture nobody had to draw.

Backtests lie, gently

Any strategy can be made to look good on history, because the rules were chosen knowing what happened. Try enough parameter combinations and something fits beautifully — and describes the noise, not a pattern.

This is overfitting, and it is the main reason promising strategies disappoint in live use. When you see "this system returned 340% backtested", the correct reaction is not interest. It is asking how many variants were tried before that one was shown to you.

We apply this to ourselves. Our model performance page reports out-of-sample accuracy — how the model did on data it had never seen. It is a less impressive number than a backtest. It is the only one that means anything. Our explainer on how AI models forecast stock trends sets out that discipline in full, including where our own model is weakest.

So why show them at all?

Because used narrowly they earn their place. The mechanics are in technical analysis essentials for the PSX, the companion to this piece: it shows how the tools are read, this one marks the fences.

• Summarising. "Above a rising 50-day average on strong volume" compresses a chart honestly.
• Divergence. Price up, momentum not — a genuine question worth asking.
• Regime detection. ADX telling you there is no trend is real information.
• Discipline. Pre-committed rules beat improvising while your money moves — though the rules that matter most concern position size and where the stop sits, not which oscillator fired.

What they cannot do is tell you what a business is worth. That question is answered by understanding the company — its earnings, its debt, its industry, its management. No oscillator substitutes for it.

Indicators describe the crowd's recent behaviour. Sometimes that is worth knowing. It is never the whole story, and on a market as thin and as policy-driven as Pakistan's, it is a smaller part of the story than the charts suggest.