It is 10.40 on an ordinary Wednesday. A stock you have been watching drops 1.2% in four minutes on no news at all. You search for a reason, find nothing, and by lunch it has recovered the whole move. Nothing happened. Something looked like it happened.
That gap — between what looks like an event and what is an event — is the first thing a beginner has to get comfortable with, because a chart shows both with equal confidence. Every technique in this track is, underneath, an attempt to separate the two.
The coin-flip chart
Take a sheet of paper and a coin. Start at 100. Flip 250 times — heads, add one; tails, subtract one — and plot the running total. That is roughly one year of daily closes generated by pure chance, and it is called a random walk.
The chart you get will have trends that last weeks. It will have levels where the line turned back three times. It will have breakouts, pullbacks, double tops, and if you look long enough, something that resembles a head and shoulders. Not one of those shapes was caused by anything, because there was nothing to cause them.
Why the eye insists anyway
Human pattern detection is tuned to fire early and often, because historically the cost of seeing a tiger that was not there was much lower than the cost of missing one that was. That setting is excellent in a forest and expensive on a chart, where a false positive costs money every time.
A batsman with a career average in the mid-thirties scores 50, 62 and 55 in three consecutive innings. The commentary declares that he has found his touch, worked something out in the nets, turned a corner. In the next three innings he makes 4, 11 and 0. Nothing about the batsman changed in either direction — a variable process produced a short run, and a story was attached to it.
Three green candles are three innings. They are draws from a distribution, and short runs happen constantly in any distribution. The story you attach — accumulation, breakout, institutions buying — is added by you, not by the data.
What raises the odds that a move means something
You cannot remove noise. You can weigh a move against it, and four questions do most of the work. None of them is a signal on its own; together they are the difference between watching a chart and reading one.
| Question | Why it separates signal from noise |
|---|---|
| How big is this move relative to the stock’s own recent range? | A ₹9 move is enormous in a stock that usually travels ₹3 a day and invisible in one that travels ₹40. Judged in rupees, every move looks arbitrary. Judged against the stock’s own recent daily range, most moves reveal themselves as ordinary. |
| Did anyone show up? | Volume is the one input not derived from price. A move on ordinary volume is the usual traffic. A move on several times normal volume means a lot of people acted at once, which is at least a fact about the world. |
| Did it happen anywhere in particular? | A 1% move in the middle of a range is weather. The same 1% move through a level that held four times over six months is a different event, because a large group of people had a decision to make there. |
| Did it survive to the close? | Intraday spikes reverse constantly. A level probed at 11 a.m. and abandoned by 3 p.m. was tested by very few participants; a level that price closes beyond has been accepted by everyone who could have sold into it all day. |
Data mining: finding signal that was never there
The second way noise fools people is subtler. Test one rule on ten years of NIFTY data and you learn something. Test two hundred rules on the same ten years and a few will look spectacular purely because you tried two hundred times. That is data mining — and it is what most "I discovered that stocks always rise on the third Friday" claims actually are.
This is not an argument against technical analysis. It is the reason the discipline exists in the form it does: define the rule before you look, apply it the same way every time, count every occurrence rather than the memorable ones, and expect a modest edge rather than a revelation.
A midcap that normally moves about ₹6 a day falls ₹7 on volume close to its 20-day average, in the middle of a two-month range. How should you read it?
Ek kaagaz lo aur 200 baar sikka uchhaalo — heads pe ek kadam upar, tails pe ek neeche — aur graph bana lo. Us graph mein bhi support milega, breakout milega, dhoondhoge toh sar-aur-kandha bhi mil jaayega. Aankh pattern nikaal hi leti hai, chahe wahan kuch ho hi na. Isiliye sawaal yeh nahi ki "pattern dikh raha hai kya" — sawaal yeh hai ki yeh harkat is stock ke rozana ke shor se badi hai ya nahi.
- Noise is real price movement that carries no information — orders arriving unevenly, not opinions about the company.
- A coin-flip chart produces trends, levels, breakouts and patterns. Shape alone is not evidence.
- Weigh every move against four things: size relative to the stock’s own range, volume, location, and whether it held to the close.
- Zooming out is the cheapest and most effective noise filter available to you.
- Testing many rules on the same data guarantees some will look brilliant by chance — define the rule first, then count every occurrence.
Mark it done to track your progress through the curriculum.
Common questions
Short, direct answers to what people ask about this topic.
- random walk meaning in stock market
- A random walk is a price series where each step is drawn by chance and carries no memory of the last one — the coin-flip chart. It matters because a random walk still produces trends lasting weeks, levels the line turns away from three times, breakouts, double tops and something resembling a head and shoulders. Those shapes are not evidence of anything, because nothing caused them.
- real price movement that carries no information about what comes next is called
- Noise. It is not an error in the data or a glitch in the feed — the trades genuinely happened. A monthly SIP inflow being deployed, an index fund rebalancing at the close, a family selling to pay a school fee, an algorithm hedging: none of that is an opinion about the company, and all of it moves price and draws candles.
- how do I tell if a price move is noise or a real signal
- Weigh it against four things: how large the move is relative to that stock’s own recent daily range, whether volume was well above its normal level, whether it happened at a level people had a decision to make at, and whether it survived to the close rather than reversing intraday. None of the four is a signal on its own, and a move that scores neutral on all four is almost always noise.
- how many occurrences before a trading rule is proven
- More than most people ever collect, and there is no clean threshold — a trading rule is a very noisy coin, so runs of three or four in a row tell you essentially nothing. Counting forty occurrences is worth doing, but for a different reason: it settles whether you can identify the pattern consistently and how common your version of it really is, not whether it has an edge.
- what is data mining bias in backtesting
- Data mining bias is what happens when you test many rules on the same history and then report the best one as a discovery. Test two hundred rules on ten years of NIFTY data and a few will look spectacular purely because you tried two hundred times. The defence is to write the rule down before you look, apply it identically every time, and count every occurrence rather than the memorable ones.