Your weekly scan returns a midcap with an unusually clean ten-year trend-following record: forty-one trades, a respectable hit rate, no catastrophic drawdown. Before sizing anything you open the exchange’s announcement archive for the symbol — a habit rather than a suspicion — and find, a little over three years ago, a scheme under which the company transferred its largest division into a separate entity and, in the same year, acquired a business roughly its own size in a different segment. Nothing was hidden and nothing was wrong with the data. The trades before that date were taken in a company that no longer exists, and the parameters that produced the record were fitted to it.
A doctor has ten years of a patient’s reports in one file. Four years ago the man had major surgery, lost an organ and changed his medication entirely. Every reading in the file is real, was properly taken and belongs to the same person. None of the pre-surgery readings can be used as a baseline for what is normal for him now, and a physician who averaged the whole file would produce a number describing nobody.
A symbol is a file, not a subject. When a company transfers out its principal business, absorbs one its own size, or has its capital rebuilt, the readings on either side of that date were taken from different bodies. The average across it describes nobody.
The assumption every lookback makes
A moving average, an average true range, a beta, a relative-strength rank, an optimised parameter and a backtest are all the same kind of object: a statistic computed over a window. Each one is valid only if the window describes one subject. This track has been careful about the window — how long, what timeframe, how many trades — and almost silent about the subject, because on a stock chart the subject appears too obvious to state. It is not. The subject is a company, and companies are rebuilt.
| The event | What actually changes | Is the earlier history usable? |
|---|---|---|
| A merger or an acquisition of comparable size | The surviving company is larger, differently financed, exposed to different customers and followed by a different set of analysts and funds | No. Volatility, beta, typical range and even the sector classification are being measured on something else |
| A demerger of the principal division | What remains is a smaller company, sometimes in a different business from the one that made the record | No, and this is the one people miss, because the symbol and the name often stay put while the substance leaves |
| A resolution plan or a capital reduction | The capital structure is rebuilt, ownership changes hands and the claim itself may have been cancelled | No, in the strongest sense. The instrument is arguably a different one |
| A change of promoter or control | Strategy, capital allocation and disclosure practice change together, and the shareholder base turns over | Treat as broken until the new arrangement has a track record of its own |
| Migration to main board from the SME platform | Trading moves to a single-share lot, so the smallest ticket anybody can write collapses; the compliance and disclosure regime tightens, and the set of buyers willing to hold it changes along with the liquidity | No. The pre-migration series was produced by a different market, not merely a smaller one |
| Entry into or exit from the derivatives segment | A whole class of participant, and a large hedging and arbitrage flow, arrives or leaves | Usable with care. Levels survive; volatility, volume and intraday behaviour shift permanently |
| Inclusion in a major index | Index funds must hold it, which changes the buyer base and the flow around every review date | Usable with care, for the same reasons |
The two tells that are actually on the chart
Two things do usually show, and neither is proof of anything on its own. The first is a permanent shift in turnover: not a spike, which happens constantly, but a level change that never comes back — average daily volume settling at three times its old level and staying there for a year. The second is a permanent shift in the volatility regime, where the typical daily range steps up or down and stays stepped. Both are consistent with a change of subject, and both are also consistent with an ordinary change in how popular a stock is. Their value is as a prompt: when the character of a series changes and does not change back, go and look for an event, because there usually is one and it is usually documented.
- 1Read the announcement archive for the symbol, newest first
Both exchanges publish every corporate announcement a listed company has made, by symbol, free, going back years. Schemes of arrangement, changes of name, changes of control and capital restructurings are all in there with their dates. Ten minutes covers a decade.
- 2Check the corporate actions record for the same period
Splits, bonuses, rights issues and demerger record dates are published as a separate list. This is also where you discover whether your chart provider has adjusted for them, which the mechanics module deals with.
- 3Pull two shareholding patterns, years apart
A large change in the share count or a change in the promoter group tells you in one glance that something structural happened, and gives you a period to search in.
- 4Set one sample start date, and write down why
The date of the most recent identity-changing event is the earliest point your data means anything. Record the event next to the date, so that six months later you are not tempted to quietly move it.
- 5Then ask whether what remains is enough
Three years of daily data is a reasonable series and a thin sample of trades for a swing system. If dating the sample honestly leaves you with too few trades to judge, the correct conclusion is that this stock cannot be used to validate the rule — not that the rule should be judged on the longer, broken sample.
- The method itself — a breakout rule tested across hundreds of names is not invalidated by one company being rebuilt
- Market-wide and sector-wide work: breadth, regime, index structure, intermarket relationships
- Anything measured on the sector or the index rather than on the individual name
- Recent structure, measured entirely after the event
- The reason you were interested, if it was a fact about the business today
- Historical support and resistance from before the event
- Beta, correlation and any pairs relationship — these are measured against a company that has changed shape
- Volatility estimates, and therefore every stop and position size derived from them
- Optimised parameters, which were fitted to the old subject
- The backtest statistics, all of them, including the ones that made you look
The ten-year record with a scheme in the middle of it
The backtest shows forty-one trades over ten years and a record you would happily trade. Three years and two months ago the company transferred out its largest division and acquired a business of similar size. Since that date the same rules would have produced eleven trades, seven of them winners, and a much shallower drawdown than the full-period figure.
A ten-year backtest on one symbol shows a strong record. Three years ago the company transferred out its principal division under a scheme and acquired a different business of similar size. What is the honest reading of the record?
Doctor ke paas das saal ki file hai, har report asli. Par chaar saal pehle bada operation hua tha aur dawaiyan poori badal gayin. Reports jhoothi nahi hain — baseline nahi ban sakti, aur poori file ka average kisi ka bhi nahi hota. Das saal ka backtest bhi aisi hi file hai agar beech mein company ne apna sabse bada division alag kar diya ho aur uske jitna hi doosra dhandha khareed liya ho. Chart bilkul seedha chalta rahega, kyunki symbol wahi hai. Pehle announcement archive khol ke shuruaat ki tareekh tay karo — uske baad jo bacha, wahi saboot hai.
- Every lookback statistic assumes the window describes one subject, and a symbol is not a subject.
- Mergers, demergers of the main division, resolution plans, control changes and main-board migration end comparability.
- None of those events prints on the price line; they live in the announcement archive.
- A permanent shift in turnover or in the volatility regime is a prompt to go looking, not proof by itself.
- Set one sample start date from a documented event, write down the reason, and refuse data before it.
Mark it done to track your progress through the curriculum.
Common questions
Short, direct answers to what people ask about this topic.
- structural break meaning in backtesting
- A structural break is a dated event after which a symbol’s earlier price history no longer describes the same subject — a merger of comparable size, a demerger of the principal division, a resolution plan, a change of control, or a migration from the SME platform to the main board. Every lookback statistic, from a moving average to an optimised parameter, assumes the window covers one subject, so a sample that straddles a break is describing an average of two companies.
- when a company moves from the SME platform to the main board its trading lot becomes
- A single share. On the SME platform trading is in lots, so the smallest ticket anybody can write is large; after migration to the main board it collapses to one share, the compliance and disclosure regime tightens, and the set of buyers willing to hold the stock changes along with its liquidity. The pre-migration series was produced by a different market, not merely a smaller one.
- where can I check if a company had a merger or demerger in the past ten years
- In the corporate announcement archive that NSE and BSE publish for every symbol, free, going back years. Schemes of arrangement, changes of name, changes of control and capital restructurings all sit there with their dates, and ten minutes covers a decade. The separate corporate actions record adds splits, bonuses, rights issues and demerger record dates.
- how far back can I backtest a single stock
- Back to the date of the most recent identity-changing event, and no further — that date is the earliest point at which the data means anything, and it should be written down next to the event that justifies it. If dating the sample honestly leaves too few trades to judge, the conclusion is that this stock cannot validate the rule, not that the rule should be judged on the longer broken sample.
- does a demerger show up on the price chart
- Usually not in any way you could read. The price line stays perfectly continuous across a merger, a demerger or a change of control, because a chart records transactions in a symbol and the symbol did not change — only the thing it refers to did. The two hints that sometimes appear are a permanent step change in average turnover or in the volatility regime, and both are prompts to go and look for a documented event rather than evidence in themselves.