Fundamental and technical analysis ask different questions

Two questions, not two answers

Fundamental analysis asks what a business is worth and compares that estimate with the price. Technical analysis asks what the price has already done and what the participants trading it appear to be doing. Different questions — so an answer to one is not a rival answer to the other.

Almost every article on this subject frames it as a contest with a winner, and that framing is inherited from people selling courses in one of them. It survives because it is easy to write. It also makes the actual disagreement invisible, which is a shame, because the actual disagreement is interesting.

Here is the shape of what follows. Each discipline gets its mechanism stated properly — what it takes in, what it produces, and what would have to be true for the answer to be worth anything. Then each gets the strongest objection against it, stated as its critics would state it rather than as a straw man. Then the two are put side by side, and the thing they actually disagree about is named.

That disagreement is not tactical. It is about what a market price is: whether it is the number most likely to be wrong, or the number that already contains everything anyone knows. Those cannot both be true, and almost nobody says out loud which one they have assumed.

One thing this article will not do is tell you which to use. That is not caution. It is that the question has an empirical answer, the evidence is contested, and a library that invented a verdict here would be worth less than one that says where the evidence stops.

What fundamental analysis is actually doing

Strip away the ratios and the spreadsheets and one claim is left. A share is a residual claim on the cash a business produces after everyone else — suppliers, staff, lenders, the tax authority — has been paid. What it is worth is what that stream of future cash is worth today — and cash arriving in ten years is worth less than the same rupees now, so it is marked down for the wait at a rate called the discount rate. Everything else in the discipline is shorthand for estimating that stream.

The work splits into two halves of very unequal difficulty. The first is reading what already happened: what the business owns and owes, set out in the balance sheet, and what it earned and at what cost, set out in the profit and loss account. This half is laborious, learnable, and mostly a matter of care.

The second half is forming a view of what happens next — how much of that profit repeats, what must be spent to make it repeat, and how long the business can hold off whoever wants to take it. This half is a forecast, and no amount of care in the first half converts it into anything else.

Watch what the forecast does to the answer. Take a business expected to produce ₹100 crore of cash a year, growing at 4%, discounted at 12% — three figures chosen to make the arithmetic visible, not because any of them is a normal value for anything. The value of a growing perpetual stream is the cash divided by the gap between those two rates: ₹100 crore over 0.08, or ₹1,250 crore. Now assume growth of 6% instead. The divisor becomes 0.06 and the answer is ₹1,667 crore.

Two percentage points on an assumption nobody can verify moved the valuation by a third. That is not a flaw in the method — it is the method working correctly on an input that is genuinely uncertain. But it means the precision of the output is borrowed, and a decimal place is not evidence.

The ratios most people meet first are compressions of this. A price-to-earnings ratio is what the market pays per rupee of last year's accounting profit; what sits inside that denominator, and what accounting policy can do to it, is the subject of PE and PB. A ratio is quicker than a forecast. It is quicker because it has assumed the forecast rather than made it.

The case against it, stated properly

The serious objection to fundamental analysis is not that the arithmetic is wrong. It is that being right about value does not oblige anyone to pay you for it, and there are three separate reasons why.

The first is that no mechanism forces a price to meet a value, and none of them supplies a date. A share is worth what the next buyer pays. If your estimate is correct and no buyer arrives at that level, the estimate sits there being correct. In a market where nobody has to transact, correct and paid are different states.

The timing is not a detail. Say your work concludes a share is worth ₹500 while it trades at ₹300 — figures chosen to show the arithmetic, not drawn from anything. If the gap closes within a year that is a 66.7% gain. If the same correct estimate takes five years to be recognised it has paid about 10.8% a year, and if it takes ten it has paid about 5.2%. Same analysis, same conclusion, three quite different outcomes, and nothing in the analysis distinguishes them.

The second objection is that the inputs are public. The accounts were published to everyone at once, the filings sit on the exchange website, and the ratios are computed by every screener in the country. Whatever the accounts say, the price already reflects the reading that the people who moved it were persuaded by. A cheap-looking ratio is therefore not something the market missed — it is the market's existing conclusion, and disagreeing with it requires knowing why it is wrong.

The third is the uncomfortable one. Over any horizon shorter than the thesis, early and wrong look identical. Both show a falling price and an unchanged argument. There is no internal test that separates them, which means the discipline can absorb any amount of adverse evidence without ever registering a refutation.

Give that failure a name, because it is the one to recognise in your own reasoning: the undated thesis. An argument with no stated horizon and no stated falsifier cannot be wrong, and a claim that cannot be wrong is not doing any work. The repair is cheap — write down, in advance, what you would expect to see by when, and what you would treat as the thesis having failed.

What technical analysis is actually reading

Technical analysis takes a different input entirely: the record of transactions. Price, quantity, time, and whatever can be derived from those three. It does not open the accounts, and its practitioners generally say so.

It is worth splitting the discipline into two layers, because they have very different standing and get argued about as though they were one thing.

The first layer is description, and it is not contested by anyone. A moving average is an average. A chart of the last sixty sessions is a chart of the last sixty sessions. That a share has traded within a band for months, or that volume on down days has exceeded volume on up days, is arithmetic performed on public data. It can be checked, it is either right or wrong, and nothing is being predicted yet.

The second layer is inference: the claim that a particular past shape carries information about the next move. This is where the disagreement lives, and it is a genuine empirical question rather than a matter of taste.

The mechanisms offered for that second layer are behavioural and structural, and they are worth taking seriously as mechanisms — which is a different thing from taking them as evidence. Participants remember what they paid, so a level where many people bought is a level where many people would like to get out even. Leveraged positions are closed by the broker rather than the owner, so a move can force transactions that have nothing to do with anyone's view. Protective stops cluster near round numbers and recent extremes, so a move through such a level triggers more transactions in the same direction. Each of those is a plausible reason a price path might carry a trace of positioning. None of them is a finding.

The sharper end of this work drops patterns altogether and reads the transactions directly. Footprint charts show how much traded at each individual price and which side initiated the trade, rather than compressing the session into four numbers. The option chain shows where positions have actually been built by strike. That is a description of participant behaviour rather than a forecast of it, and describing it accurately is a real skill with an honest ceiling: positioning is not intention.

The case against it, stated properly

The serious objection is not that charts are superstition. It is that the second layer — the inference — is remarkably hard to test, and the ways it fails testing are subtle.

Start with the plainest one. A pattern is a description of the past, and it is named after it has finished. The examples that teach it are the ones that resolved the way the name says; the ones that did not resolve were never labelled, because the label requires the resolution. The exceptions are invisible by construction, and a rule learned from a filtered sample is a rule about the filter.

Second, most pattern definitions are not precise enough to be tested by someone other than their author. A testable definition would have to say how far a shoulder may sit from level, and how long a consolidation may run before it stops counting as one; most definitions say neither. If two competent people can look at the same chart and disagree about whether the pattern is present, then no dataset can settle whether the pattern works, because no dataset can be constructed without first settling the disagreement.

Third, and this is the one that catches careful people: search costs nothing here. Test a single rule at a threshold that noise would clear one time in twenty, and a pass is mild evidence. Test a hundred variations against the same history at the same threshold and roughly five will clear it on nothing at all — 100 × 0.05 = 5 — and those five are the ones that get written up as a strategy. Nobody has cheated. The arithmetic of searching does it on its own. The discipline of testing a rule chosen in advance, on data it was not selected from, is the subject of honest backtesting, and it is mostly a discipline of refusing to look twice.

Fourth, the data is public and the computation is trivial. Anything simple enough to be taught in a weekend can be computed by everyone at once, which is the same objection made two sections ago against a cheap-looking ratio. A rule's being widely known is not proof it stopped working. It does mean the argument for why it still works cannot be that other people have not noticed.

Then there is the reply practitioners give — that levels matter precisely because enough people watch them, so the belief creates the effect. Take that argument seriously and notice it runs both ways. If a level works because it is crowded, then the crowd is the mechanism, and a crowd that grows past some point is transacting against itself. Self-fulfilling is also self-limiting, and the same sentence cannot be used only when it flatters.

The mirror of the undated thesis lives here too, and it deserves its own name: the clean-setup defence. When the rule fires and loses, the loss is attributed to the setup not having been clean, the conditions not having been right, or the trader not having followed it properly. Each of those may be true in a given case. As a habit it makes the rule unfalsifiable, and it is the same escape the fundamental analyst uses when a thesis is described as merely early.

The two, side by side

Set the mechanics out in one place and the symmetry becomes visible — not a symmetry of quality, but of structure. Each is precisely silent about what the other looks at, and each fails in the way its own strength implies.

Read the last row as the price of admission for each. Neither is a small requirement, and neither is absurd. A reader who can state their own approach's version of that row is doing better than most of the material written about either.

The disagreement underneath, which nobody names

They disagree about what a market price is — not about indicators, ratios or rigour. Look again at the fourth row of the table. To the fundamental analyst, the market price is the error term — the quantity being measured against an independent estimate of value, and therefore the thing that may be wrong. To the technical analyst, the market price is the signal — the aggregated conclusion of every participant, including everyone who has already done the fundamental work, and therefore the best available summary of what is known.

Same number, opposite epistemic role. One discipline exists because price can be wrong; the other exists because price is the most informative thing available. That is a real disagreement about how markets process information, and technique cannot settle it.

Notice what it implies for the popular compromise. If price already contains everything known, an independent valuation cannot be an edge; if price is frequently wrong, its recent path is a record of that error rather than a guide. Holding both positions at full strength is not sophistication. It is an unnoticed contradiction, and the way it gets resolved is worth watching for in your own reasoning: whichever of the two currently agrees with the position gets the casting vote, and the other is filed as noise.

What would settle it, in principle, is evidence: a rule specified in advance, applied to data it was not chosen from, measured after transaction costs, on a sample long enough that the result is not one market regime wearing a disguise. That work is possible. Whether either approach has predictive content once it is done properly is an empirical question, and it is not one this article settles: we have not run that test, and a second-hand summary of somebody else's is not evidence. What we can say without evidence is what each is betting on — which is what the table above sets out, and it is the more useful thing to know.

One asymmetry is worth stating because it is mechanical rather than empirical, and it cuts against both. Every participant sees the same prices and the same filings. A method that works because of information nobody else has is not available to a retail investor in either discipline, so both are, in practice, claims about processing public information better — not about having more of it.

Where they meet, and the two mistakes that follow

The usual compromise is stated as a division of labour: fundamentals decide what, technicals decide when. As a sentence it is fine. As a practice it has two characteristic failures, and both deserve a name.

The first is the timeframe mismatch. A valuation argument speaks to years, because that is the horizon over which the cash it is about arrives. A chart of recent sessions speaks to the period it covers. Buy on a five-year argument and then abandon the position after a fortnight of falling prices and you have not combined the two — you have used one to enter and a different, faster one to exit, which means the thesis was never tested at all. The exit rule chose the horizon, whatever the entry note said.

The second is confirmation by search. You form a view, then go looking at charts until one agrees with it, and count that as independent support. It is not: you selected it because it agreed. This is the ordinary machinery of how people evaluate evidence working exactly as it normally does, which is why noticing it in yourself is hard and noticing it in someone else is easy.

There is a version that avoids both, and it is not complicated. Decide in advance which question each tool is answering, on what horizon, and what result would count as that tool having been wrong. Then, when they disagree, treat it as an unresolved question rather than a tie broken by whichever is currently more comfortable. Two methods pointing the same way is one observation more than one method pointing that way. It is not proof, and it is not two.

The trade-off in running both is real and is rarely stated. Every additional criterion reduces the number of situations that qualify, which is the point, and also increases the number of judgements between you and a decision, which is not. More filters is not more rigour past the point where you can no longer say which filter did the work.

Doing either one on data instead of arguing about it

Both disciplines improve in the same way: by being specified precisely enough that they could turn out to be wrong. That is mostly a matter of writing the rule down before looking, and then computing it over a universe rather than over the examples that came to mind.

FNOTrader's Stocks app screens across roughly 2,390 stocks and 17 NSE sector and size indices on reported fundamentals and on price-derived measures in the same query, so a filter can be stated once and applied to everything rather than to a shortlist. How a screen quietly answers a different question than the one you asked is set out in the screener guide. On the transaction side, the footprint and market-profile views reconstruct what traded at each price, and the option chain reports positioning by strike; both are covered in their own articles linked above.

What no tool does is choose between the two questions, or convert either answer into a decision. It computes the same thing the same way across a universe, which is what makes two results comparable, and it shows the inputs next to the output. The reading stays yours.

FNOTrader is not a SEBI-registered investment adviser or research analyst, does not recommend shares, and nothing here is a view on any company, any price or any method.

Common questions

Continue reading

More in Stocks & Screening · App: Stocks · Definitions: glossary · Free tools: calculators · All: every article