The reckoning

The first nine posts covered candlesticks, levels, trend lines, averages, volume and oscillators. Now the reckoning — not as a disclaimer bolted on at the end, but because a series that taught you those tools and skipped this would have been selling something.

Everything below is drawn from the same Britannia dataset the rest of the series used. None of it is hypothetical. A second module follows this post — volatility, relative strength, multiple timeframes and backtesting — and it keeps the same rule: every signal gets checked against what actually happened.

1. It cannot see anything that isn’t in the price

This is the structural limit, and no indicator gets around it.

A price series knows nothing about whether the company is solvent, whether its auditors resigned, whether receivables are being recognised aggressively, or whether the promoter has pledged their shareholding. The fundamental toolkit exists precisely because those questions cannot be answered from a chart.

When a company fails, the chart shows the failure happening. It does not show it coming, and afterwards everyone points to the chart and says it was obvious. The OCF/PAT ratio can raise a question about earnings quality. A moving average cannot, ever, in principle.

2. Signals are far rarer, or far noisier, than the tutorials suggest

The tutorials show you a signal and its outcome. They don’t show you the base rates. From this dataset:

Indicator What two years actually contained
50/200 crossover 1 golden cross, 0 death crosses observed (the 50-day was already below the 200-day when the 200-day first existed)
MACD signal crossover 31 signals — one every ~15 trading days
RSI above 70 42 days

Look at the first and second rows together. The golden cross is so rare that two years of data yields a single instance — from which nothing whatsoever can be concluded. The MACD crossover is so common that most of its signals are noise, clustered exactly in the sideways periods where they’re least informative.

Neither is a usable evidence base from one chart. One is too rare to test; the other is too noisy to trust. That’s not a criticism of these particular indicators — it’s the general shape of the problem.

3. The same signal produces opposite outcomes

The RSI table from that post is worth repeating, because it is the single most honest thing in this series:

Signal Date RSI Change over the next 63 sessions
Overbought 2024-07-09 70.4 +7.6%
Overbought 2024-09-20 72.0 −24.3%
Oversold 2024-11-12 23.2 −2.5%

Every outcome is measured over the same fixed window — about three months — not to whichever peak or trough came later.

Two nearly identical overbought readings, ten weeks apart, on the same stock. One preceded a 7.6% rise. The other preceded a 24.3% fall. Nothing available at the time distinguished them.

4. It is trivially easy to fool yourself

The chart patterns post demonstrated this rather than asserting it. A textbook double top — two peaks 0.15% apart — dissolved once every swing high in the window was listed: six of them, in a 3.1% band, with the tallest occurring before both chosen peaks and another arriving after them, before the pattern had even confirmed.

Nobody set out to deceive anyone there. That’s what makes it worth dwelling on. Selecting the points that fit is what pattern recognition does, and the only defence is procedural: enumerate everything first, decide the criteria before you look, and write down what would prove you wrong.

5. Every “objective” indicator hides a choice

Technical indicators feel objective because they’re arithmetic. They aren’t, because someone chose the parameters:

  • Why 14 periods for RSI, not 9 or 21?
  • Why 12/26/9 for MACD?
  • Why a 50-day and 200-day average, rather than 40 and 180?
  • How many bars either side define a “swing high”?
  • Does a trend line break on one close beyond it, or two?

Every one of those changes the signals. And the trend lines post showed how much can ride on it: the January 2025 downtrend break happened by ₹2.40 on a single close. A slightly different rule, and there was no signal that day at all.

Which brings up the deeper problem. Search enough parameter combinations against data you already have and something will look excellent by chance alone. That’s overfitting, and it’s the reason strategies that backtest beautifully so often disappoint afterwards. If a rule’s performance collapses when you change 14 to 15, you found a feature of that dataset, not of markets.

6. One stock over two years proves nothing

This series has been careful about this and it bears restating plainly. Everything here came from 495 bars of one large-cap Indian stock over two years. That is an illustration of mechanism — how an indicator is computed and what it looked like when it fired. It is not evidence about whether any of it works.

Real evidence would need thousands of instances across many stocks, sectors, market conditions and decades, with trading costs included and the rules fixed in advance. Where that work exists, the results are modest. Park and Irwin’s 2007 survey of the research (Journal of Economic Surveys) found that many studies reporting profitable technical rules were weakened by data snooping, rules picked after the fact, or understated costs. And Sullivan, Timmermann and White (1999) found that the best moving-average rules from an earlier century-long study stopped working in the decade after it.

Be suspicious of anyone who shows you one chart. Including this one.

7. If a simple rule worked, it would stop working

The first post left a tension open. Technical analysis assumes the price already reflects everything — and then studies past prices to guess the next one. Economists have a name for the relevant idea: weak-form efficiency (Fama, 1970). It says past prices are already reflected in today’s price, so no rule built only on past prices should beat the market after costs, except by luck.

You don’t have to believe markets are perfectly efficient to feel the force of the argument. If a simple chart rule reliably made money, traders would pile into it — buying a little earlier, selling a little earlier — and in doing so trade the edge away. So the rule printed in every textbook is the one least likely to still work. That doesn’t settle the debate; some effects, like momentum (Jegadeesh and Titman, 1993), have been studied for decades and are still argued over. But it explains why the edges researchers do find tend to be small, short-lived and eaten by costs.

So is any of it worth knowing?

Yes — held at the right level of confidence.

What it’s reasonably good for:

  • A compact description of what happened. A candlestick chart genuinely packs more information into less space than any table of prices.
  • Identifying where price has previously reacted. The ₹6,145–₹6,470 band was tested seven times over seventeen months. That is a real, observable fact about this stock, not an interpretation.
  • Imposing discipline. Deciding in advance what would count as a break, and writing it down, is a genuinely useful habit — arguably more valuable than any signal it generates.
  • Understanding what other participants are watching. A lot of people watch the 200-day average. That alone gives it some effect, regardless of whether it has predictive merit.

What it cannot do:

  • Tell you what happens next
  • Reveal anything not already expressed in price and volume
  • Assign a probability to any outcome
  • Distinguish, at the time, a signal that will work from one that won’t
  • Substitute for knowing what the business is and whether it can pay its debts

The honest position

Technical analysis is a language for describing price behaviour. As a description, it’s genuinely useful — more compact and more precise than talking about charts in prose.

As prediction, the evidence is much weaker than its confident vocabulary implies, and the vocabulary is part of the problem. “Overbought” sounds like a fact about value; it’s a statement about the last fourteen sessions. “Confirmed breakout” sounds like something settled; it’s a threshold someone picked. Words like these do a lot of quiet persuading.

The most useful thing this series can leave you with isn’t a setup. It’s a habit: before accepting any signal, ask how many times that signal occurred in the same data and what happened on the other occasions. Ask what you’d have to ignore for the pattern to hold. Ask what would change your mind.

Applied honestly, that habit will disqualify most of what gets published about charts — including, quite often, the chart in front of you.

Takeaway: Technical analysis describes what price has done, compactly and sometimes usefully, but it can’t see anything outside the price, and the same reading routinely precedes opposite outcomes. Any simple rule that did work reliably would tend to get traded away once enough people used it. Treat it as a vocabulary for describing markets rather than a method for predicting them — and be most sceptical of the chart that fits your existing view best.