How to Use FII and DII Data Without Over-Reading It

Use FII and DII data as slow moving context about who owns the Indian market and how that ownership is shifting, not as a daily signal. The useful unit is a rolling sum over months, read next to price, valuation, and whatever else changed in the same window. The daily net print, which is what most commentary quotes, is the least informative form of the series.

This article is about method: how to frame a flow question so it can be answered, how to pick the right dataset, and how to test a flow claim without fooling yourself. If you want the underlying definitions and publication structure first, start with FII and DII flows explained.

Start by fixing the question

Almost every bad piece of flow analysis begins with a vague question. “What are FIIs doing” cannot be answered because it does not specify a horizon, a market segment, or a comparison. Three sharper questions can be answered, and they need different data.

Question one: has the ownership mix shifted? This is a stock question, about holdings rather than trades. It is answered with custody based depository data and with company level shareholding pattern disclosures, over quarters and years.

Question two: which side absorbed supply in a given period? This is a flow question about a defined window. It is answered with exchange cash market activity, summed over the window, for both institutional groups at once.

Question three: is a specific claim about flows and returns true? This is a research question, and it needs a testable statement, a dataset with correct timestamps, and an out-of-sample check. Most flow commentary never gets this far.

Writing the question down before opening the data is not a formality. It determines which of the two incompatible datasets you should be using, and it stops you from quietly changing the question when the first answer is boring.

Choose the dataset deliberately

The exchange series and the depository series both describe foreign institutional activity in India and they do not agree, by construction.

The exchange series counts on-exchange cash segment trades, tagged by client category, published per exchange on a daily basis, provisional first and settled after. It is the right choice for questions about trading activity in a defined window. It excludes derivatives, excludes primary market subscription, and excludes off-market transfers.

The depository series is built from custody records, so it captures changes in what foreign portfolio investors actually hold, including allotments in initial public offerings and transfers that never crossed the order book. It is published net, with a lag, and generally with more structure such as an equity and debt split and sector groupings. It is the right choice for questions about allocation and holdings.

A third surface, the exchange participant-wise open interest data in the derivatives segment, is what you need if the question involves positioning rather than cash buying. Reading only the cash file and concluding that an institution is bullish is a common and avoidable error, because the same desk may express the opposite view in index futures.

Never plot the two flow series on one chart as if they were one measurement. If a claim only works when the datasets are spliced, the claim is an artifact of the splice.

Aggregate before you interpret

The single most effective discipline in flow analysis is to stop looking at days.

• Use rolling sums. A twenty-session or three-month rolling net is a far better description of an allocation stance than any single print. Allocation decisions at large institutions are made and executed over weeks, not in one session.
• Scale the number. An absolute rupee figure is not comparable across years, because the market it flows into has grown. Expressing flow relative to market turnover, or to free float market capitalisation, makes different periods comparable. Free float is the base most index construction uses, for reasons covered in free float market cap explained.
• Read both groups on one axis. Because domestic institutional flow is substantially mechanical, driven by recurring savings inflows, and foreign flow is substantially discretionary, the informative pattern is usually the divergence between them and how long it persists.
• Separate gross from net. Gross activity measures engagement. Net measures direction. A quiet net figure on heavy gross turnover is a completely different market from a quiet net on thin turnover.

Test flow claims the way you would test any strategy

Suppose someone asserts that heavy foreign selling over a month has historically been followed by something. That is a testable statement, and testing it properly requires the same machinery as any other backtest.

Respect the publication lag. The flow figure for a session is available after that session, and the settled figure later still. A test that lets a rule act on the same day’s flow is using a number that did not exist at the decision point. This is textbook lookahead bias, and it is easy to introduce accidentally because the file is stamped with the trading date rather than the publication time. A point-in-time discipline, described in why point-in-time data matters, is what keeps the timestamps honest.

Count the parameters. A flow rule has a lookback window, a threshold, a holding period, and often a filter. Four free choices over a market history of a few hundred non-overlapping months will produce impressive-looking rules from noise alone. This is the mechanism explained in what is overfitting in backtesting.

Hold data back. Fit on one period, evaluate on another you have not looked at. If a flow rule only works in the sample where it was found, you found a pattern in that sample and nothing more. The reasoning is set out in in sample vs out of sample testing.

Check the regime. The relative size of domestic and foreign institutional participation in India has changed materially over the past two decades. A relationship estimated on an era when one group dominated may not describe a market where the balance is different. A rule fitted across a structural break is fitted to two different markets averaged together.

Ask what else was happening. Index inclusion events, large rebalances, expiry weeks, and currency moves all generate flow that carries no view about Indian companies. If your result is concentrated in those windows, you have found a calendar effect wearing a flow costume.

What it does not tell you

Flow data is over-read more than almost any other public series in Indian markets, so the limits deserve their own list.

• It is descriptive, not causal. Prices and flows are two readings of the same transactions. “The market fell because institutions sold” is a restatement of the day, not an explanation of it.
• It contains no reasoning. Rebalancing, redemption pressure, a currency hedge, and a considered valuation view produce identical rows.
• It is an aggregate over disagreeing parties. Hundreds of registered entities with opposite mandates net down to one figure. Zero net can mean total agreement or violent two-way disagreement.
• It rarely reaches stock level. Public flow data is a market and, with a lag, a sector measure. Company level institutional ownership comes from quarterly shareholding disclosures, which are a different, slower, and much more precise instrument.
• It says nothing about business quality. No flow series tells you whether a company converts profit into cash or earns a return above its cost of capital.
• It is not a timing tool, and this article does not offer one. The honest use is context.

Held to that standard, institutional flow data earns a real place in a research process. It describes the ownership weather: which pools of capital have been adding, which have been trimming, and how durable that has been. That is worth knowing. It is simply not the same thing as knowing what to do.

Related reading

• Portfolio metrics explained: the hub for the risk, return, and market analytics in this series.
• FII and DII flows explained: who the two groups are and how the data is published.
• Market breadth indicators: the companion market-level health check and its failure modes.
• Common backtesting mistakes: the errors that make any rule, flow-based or not, look better than it was.
• Seasonality analysis in Indian markets: the same data-mining risk in a different costume.

This article is educational. Altys Labs is not a registered research analyst or investment adviser, and nothing here is investment advice or a recommendation to buy, sell, or hold any security.