In the first half of 2025, Bosch Limited was a mid-cap stock. In the second half, it was a large-cap stock. Nothing about Bosch had changed. It still made fuel injection systems and power tools, it still had the same customers, and its revenue was what it had been the quarter before. What changed was that the Association of Mutual Funds in India, which every six months ranks the top 500 listed Indian companies by market capitalisation and publishes the order, moved Bosch up one seat. On one side of the line, Bosch is the kind of stock a mid-cap fund can count toward its compliance floor. On the other side, it is not. The stock is the same stock. In the mutual fund rulebook, it is a different species.
Bosch does this sort of thing more than most. Its market cap sits close enough to the boundary between the top 100 and the next 150 that small movements in other companies, bigger ones getting slightly smaller, smaller ones getting slightly bigger, push it back and forth. Its business does not need to move for the label to move. The label moves on its own.
For most investors this is invisible, a line item in a list nobody reads. For some funds, it is not. Every equity mutual fund in India has, since October 2017, been required to fit into a named category with a compliance floor. Large-cap funds must keep at least 80 percent of their equity portfolio in stocks from the top 100 of the AMFI list. Mid-cap funds must keep at least 65 percent in stocks ranked 101 to 250. Small-cap funds, 65 percent in stocks ranked 251 and below. When AMFI publishes its updated list on 30 June and 31 December each year, some stocks cross a line. The mid-cap fund that owned Bosch in June 2025 did not sell Bosch in July. It did stop counting Bosch toward its 65 percent. The holding stayed. The label did the work.
This piece is about what happens at that seam between lanes, where an administrative list becomes, twice a year, an operational event. The effect is modest. The evidence is partial. Nothing in it suggests misconduct, or forced selling, or that anyone is front-running anybody. What the data suggests, in a small and local sample, is that reclassification days are not neutral, that some funds trade around them, that some stocks see elevated trading activity in the days after, and that a simple change to how the rule draws its boundaries would have removed a meaningful share of the churn in reconstructed history. Which makes this, I think, an unfinished design problem rather than a scandal.
I. THE RULE AND WHAT IT WAS TRYING TO FIX
Before October 2017, mutual fund categorisation in India was mostly a marketing decision. A fund could call itself a large-cap fund and hold a lot of mid-caps. It could call itself a multi-cap fund and do more or less anything. There were roughly two thousand open-ended schemes in the country, many of them distinguishable only by the font on their brochures, and investors had no reliable way to tell what they were actually buying. If you handed your money to a "large-cap growth" fund, you might find that a quarter of it was sitting in mid-caps that had done well recently, because those mid-caps were juicing the fund's one-year returns, and nobody was going to stop you finding out later.
SEBI's categorisation circular in October 2017 cleaned this up. Every open-ended equity scheme had to pick a lane and stay in it. Five broad categories were defined, each with investment characteristics and a minimum allocation floor. A large-cap fund had to hold at least 80 percent in large-caps. A mid-cap fund needed 65 percent in mid-caps. A small-cap fund needed 65 percent in small-caps. Hybrid and thematic categories had their own rules. Calling yourself a large-cap fund stopped being a marketing choice and became a compliance one, enforced by the regulator, audited periodically, and attached to a penalty if you broke it.
Most people who watched the circular arrive thought it was a good idea, and most of the time, it still is. Investors now know roughly what they are buying. Comparisons between funds in the same category make more sense. The fund industry is harder to mislead with. The rule did useful work.
What the rule also did, as a condition of working, was require somebody to decide which stocks were which. Someone had to keep the list. That job fell to AMFI. The mechanism is straightforward. Every six months, AMFI takes the average full market capitalisation of listed Indian companies over the previous six months, ranks them, and publishes the order. The top 100 are large-cap. The next 150 are mid-cap. Everything from 251 to 500 is small-cap. Funds get a window of about a month to adjust, after which the new categorisation is binding.
A classification rule of this shape has a known property. When you force a continuous variable into discrete categories by drawing hard lines at fixed ranks, things that live near the lines spend a disproportionate share of their time crossing them. Bosch is one of those things. So is Cummins India. So is Lupin. So are a few dozen others in any given year. Twice annually, one or several of them cross, and for the funds that hold them in size and sit close to their compliance floor, that crossing becomes an event worth paying attention to. This is not a flaw in the rule's intent. It is a feature of how the rule is drawn.
II. THE MECHANISM, SLOWLY
Here is how the pressure works at the scheme level, which is where it lives.
A mid-cap fund with Rs 50,000 crore in AUM and a 65 percent mid-cap floor has to keep at least Rs 32,500 crore in stocks from the 101-to-250 band of the AMFI list at any given time. In practice, funds do not hug the floor. They keep a cushion. Some funds keep a ten-point cushion and sit at 75 percent. Some keep a three-point cushion and sit at 68. The size of the cushion depends on the manager, the mandate, recent flows, market conditions, and a dozen other things. On most days, it does not matter.
On reclassification days, the cushion matters. If AMFI's new list tells you that three percentage points of your portfolio just moved from mid-cap to large-cap or small-cap, you are three points closer to your floor. If your cushion was ten, you still have seven, and you can wait. If your cushion was three, you have zero, and you need to decide what to do. You can let the market move in your favour. You can buy more of the mid-caps you already like. You can trim the reclassified names and replace them. What you probably will not do is nothing, because the floor is binding.
Notice what the rule has just done to the manager's priorities, without anyone breaking anything. A stock that was in the fund's bucket yesterday is no longer in the bucket today. Holding that stock now creates a cost the manager did not ask for. The business has not changed. The case for owning the stock has not changed. What has changed is that the stock, by sitting in the manager's portfolio, is now pulling the portfolio closer to a floor the manager is required to honour. There is no obligation to sell. There is, however, a reason.
Now extend this across every scheme in the country that happens to sit close to its floor on a reclassification date, and you get something that stops looking like a coincidence of independent portfolio decisions and starts looking like a mild, recurring, partly predictable pressure at the category edge. Some of that pressure gets absorbed by patience and flows. Some of it shows up as trading.
This shape of problem, where a hard regulatory line creates incentives at the boundary, is not unique to mutual funds. It appears anywhere a continuous thing like market cap, or credit score, or income, is forced into discrete administrative categories for a good reason. The question is usually not whether to have the categories. It is how to draw the line so that the thing being sorted does not spend all its time bouncing across it.
III. ONE FUND, ONE RECLASSIFICATION, ONE INVESTOR
To see what this looks like inside an actual portfolio, take HDFC Mid Cap Opportunities Fund. HDFC Mid Cap is the largest actively managed mid-cap scheme in India. It had roughly Rs 84,000 crore in AUM at the end of June 2025. It is held by several million unitholders, run by a fund manager with a long track record, and as of early 2026 remains closed to large lump-sum subscriptions because it has got, to use an industry word, too big.
On 30 June 2025, AMFI published a reclassification that touched HDFC Mid Cap in both directions. Five stocks that had been mid-caps moved to large-cap: Indian Hotels, Indraprastha Gas, Jindal Steel & Power, Solar Industries, and Union Bank. Four stocks previously outside the mid-cap band entered it: Bosch (yes, that Bosch), Cholamandalam Financial Holdings, Cummins India, and Dabur India. From the fund's disclosures, the arithmetic on a hypothetical Rs 10 lakh investor position on that date looks like this.
Total exposure to stocks involved in the reclassification, as of the event-month snapshot: around Rs 1,21,700. Of that, the slice held in names leaving the mid-cap bucket: about Rs 85,700. The slice held in names entering: about Rs 36,000. Indian Hotels alone accounted for around Rs 19,400 of the Rs 10 lakh holding. Solar Industries, around Rs 18,400.
In the months that followed, both Indian Hotels and Solar show observable reductions in HDFC Mid Cap's reported unit holdings. Indian Hotels drops from roughly 2.15 crore units to 1.68 crore within a single reporting month. Solar drops more sharply, from about 8.77 lakh units to 72,000 in the first month after, and to zero in the second. These are the numbers. The fund is not obligated to tell anyone why it cut, and it did not. The movement is consistent with pressure-driven rebalancing, with a change of view on the stocks, with flows in or out of the fund, with tax considerations, or with any combination of those working in different directions at the same time. What the data shows is that the mechanism was available, that the rule supplied a reason, and that the positions moved.
For the investor who sold nothing and did nothing, the effect of the reclassification was that one hundred thousand rupees or so of exposure got rearranged inside the fund without consultation, on a schedule set by an industry list, in response to ranking movements in companies unrelated to the fund's thesis. This is not a bad outcome. It may, in the fund manager's judgement, have been the right outcome for reasons that had little to do with AMFI. It is, however, an outcome that the investor would not have got to by buying and holding Indian Hotels themselves, and which nobody explained beforehand because nobody has a reason to.
IV. THE WIDER PATTERN
HDFC Mid Cap is a useful teaching case because the numbers are legible and the reclassification was concentrated, but the pattern does not live inside one fund. If it did, it would be a curiosity. The question is whether anything similar shows up elsewhere.
On the 31 December 2025 reclassification, the stocks with the highest cumulative pressure across schemes were not, for the most part, the ones that would make a good magazine opener. They were Cummins India moving from mid-cap to large-cap, Lupin moving from large-cap to mid-cap, NLC India moving from mid-cap to small-cap, Apar Industries moving from small-cap to mid-cap, and Mankind Pharma moving from large-cap to mid-cap. Each of these stocks crossed a category line, each was held by multiple schemes at multiple AMCs, and in several cases, the holdings sat inside schemes close to their compliance floor. For Cummins, aggregate pre-event market value across the observable schemes summed to around Rs 2,673 crore, with at least one scheme carrying more than two percent of NAV in the stock.
Whether that pressure translated into actual unit reductions depended on which AMC and which scheme. For Cummins, in the observable sample, it mostly did not. The positions stayed. For Apar, which crossed in the other direction, it partly did. One Nippon small-cap scheme reduced its Apar holding by about two percent of units in the month after the reclassification, a smaller move than HDFC's Solar exit, visible, and consistent with a fund that had less slack than average.
The cluster is useful because it scrubs the anecdote out of the argument. Pressure is not the same thing as action. Some stocks cross the line and nothing happens, because the funds that own them have room to wait. Others cross the line and something happens, and the something happens quietly, spread across a few schemes, and never gets a press release because, from the fund's point of view, nothing particularly interesting has occurred. The reclassification is routine. The trade is routine. The rule is the rule.
V. IS THIS JUST HOW FUNDS TRADE ANYWAY
A reasonable reader, reading this, might be wondering something obvious. Mutual funds trade all the time. Positions change every month. If I tell you that five out of twenty-seven unit-observable, reclassified scheme-stock rows showed a visible cut in the months after, the next honest question is whether this is higher than the baseline for stocks the funds held that were not reclassified.
Inside the same schemes and the same months, unit reductions showed up in about 18.5 percent of reclassified rows and about 13.6 percent of non-reclassified ones. The gap is roughly five percentage points. It is not dramatic, and it is not zero. It runs in the direction the mechanism would predict without doing the more ambitious work of proving cause. At HDFC the gap is closer to seven points. At Nippon it is smaller. The unevenness is useful. It says the effect is not uniform, which is what you would expect if it were driven partly by where each scheme sits relative to its floor rather than by some general force acting on all reclassified stocks equally. It also limits how far the finding travels. This is a pattern in a sample, not a claim about the industry.
VI. WHAT SHOWS UP ON THE TAPE
The question a reader will next ask is whether any of this is visible to someone watching prices and volumes. The short answer is that it sometimes is, and that when it is, the signal lives in activity more than in direction.
Of the twenty-five highest-pressure reclassification events in this dataset, around half show elevated trading volumes in the implementation window once you adjust for what comparable non-reclassified stocks were doing in the same days. Some of those events involved observed unit cuts, some did not. Solar Industries and Indian Hotels both show above-average trading activity in the weeks after 30 June 2025, which is at least consistent with other funds behaving the way HDFC did. Cummins, despite significant aggregate pressure, shows nothing much. Price effects are small, mixed in sign, and indistinguishable from noise for most events. If you came to this material hoping for a trade, you will leave disappointed. The activity, where it exists, is in how much is changing hands, not where the stock ends up.
VII. SCALE, AND WHY IT MATTERS MORE NOW
How much money actually sits near the edge of a category floor on a reclassification day? In the local sample, more than a small amount.
Four scheme-dates in this data sit within five percentage points of their relevant compliance floor on a reclassification date. Their combined reported AUM is roughly Rs 2,87,000 crore. Three of those four are within a single percentage point of the floor. Their combined AUM is roughly Rs 1,94,000 crore.
Rs 2,87,000cr
Combined AUM of near-floor schemes · sample 40%
Reclassification churn removed by buffer-band rule
The largest near-floor scheme-date in the sample is HDFC Mid Cap at the end of December 2025, with about Rs 92,700 crore. The second is the same fund six months earlier at about Rs 84,100 crore. The third is Nippon India Small Cap at about Rs 68,300 crore. The fourth is Nippon India Growth Mid Cap at around Rs 42,100 crore. None of these are small funds. Several of them are among the largest in their categories.
Two things about this number. It is a local-sample figure, drawn from six AMCs, not an industry total. And it does not describe money that was "forced to trade." The rule allows for considerable latitude. What the number describes is the size of the schemes, in this sample, that had the least room to manoeuvre on a given reclassification day, and that therefore faced the most pressure. Which is the piece's answer to the natural next question, which is why this matters more now than it did when SEBI drew the rule.
When the 2017 categorisation circular was issued, Indian equity mutual fund AUM was around Rs 7 lakh crore. As of early 2026, it has crossed Rs 30 lakh crore, a more than fourfold increase over roughly eight years, most of it driven by retail money arriving in systematic investment plans. The rule has not changed. The thing the rule is regulating is now four times bigger. Edge-case friction that was minor at Rs 7 lakh crore of AUM is less minor at Rs 30 lakh. Twice a year, on 30 June and 31 December, somewhere between a handful and a few dozen stocks cross a line maintained by a ranking list, and somewhere between a handful and several dozen schemes, collectively holding the savings of millions of retail investors, have to decide what to do about it. The question of how well the line is drawn is, at that scale, not a trivial one.
VIII. A FIX THAT IS NOT A REFORM
There is a small design change that would, in reconstructed history, take a meaningful fraction of the churn out of the system without touching anything about what the rule is trying to do.
It is called hysteresis, which is an engineering word for a feature familiar to anyone who has used a thermostat. A thermostat set to 22 degrees does not switch on the moment the room drops to 21.99 and off again at 22.01. It has a buffer zone, usually around a degree. It waits until the room is clearly below 22 to switch on, and clearly above it to switch off. This is not a compromise with the setting. It is a recognition that the thing being measured wobbles in the short run, and that reacting to every small movement produces more churn than it prevents.
Applied to AMFI's categorisation, hysteresis would look roughly like this. A stock gets reclassified only when it has clearly crossed a category boundary and stayed on the new side for more than one consecutive update. A small buffer zone around each rank cutoff, within which stocks retain their previous classification, would absorb most of the oscillation that currently gets converted into semiannual reclassification events.
Reconstructing the AMFI updates with a simple buffer-band rule applied across recent history, the number of reclassifications drops from 381 to 279. A hundred fewer border crossings over the period. Across observable scheme-months, aggregate absolute bucket shock, the sum of how much each scheme's mid-cap or large-cap or small-cap weight changes between the old and new maps, drops from 21.45 percentage points to 13.23. Somewhere close to 40 percent of the churn, removed by a rule change that does not relax anything about the underlying categories. The implementation window does not change. The compliance floors do not change. What changes is which stocks cross the line, and how often they cross back.
The hysteresis idea is not novel. Morningstar, the American research firm that has been classifying funds for decades, uses buffer zones around its style-box boundaries for exactly this reason. The difference between Morningstar's boxes and AMFI's list is that Morningstar's are descriptive. A stock drifting in and out of the large-blend box is mildly annoying to an analyst and nothing else, because no compliance floor hangs on it. AMFI's list, because of the 2017 architecture, has operational consequences. Which means the argument for hysteresis in the Indian context is, if anything, stronger than it is in the descriptive context where the idea is already standard.
IX. ON CLASSIFICATION
Every system that imposes hard boundaries on a continuous thing has to decide how to handle what happens at the edge. Credit scores draw lines at 650 and 750 that determine whether a loan gets approved and at what rate. Insurance classifies risk in ways that can flip with a small change in a medical report. Tax systems draw lines between brackets. Regulatory definitions separate small businesses from medium ones based on turnover thresholds that the business itself may not find meaningful. In each case, a continuous variable is being forced into discrete categories because discrete categories are easier to administer and easier to communicate.
The problem with hard boundaries is that things near the boundary spend a lot of their time crossing it. A credit score of 649 is almost indistinguishable from a credit score of 651. The consequences of being one rather than the other can be large. Most serious systems fix this with some combination of smoothing, averaging, staged transitions, or explicit buffer bands. Sudden cliff edges tend to get designed out, because a rule that treats a thing very differently depending on which side of a line it happens to be on creates an incentive for the thing to sit on the line, or to bounce, or to be nudged from one side to the other by actors who care about the consequence.
The 2017 categorisation framework is a reasonably well-designed rule that did most of its work and left an unfinished edge. The work was cleaning up how funds label themselves. The work got done. What remains is a boundary treatment that was not designed for the scale of money now sitting inside the architecture, and that produces, twice a year, more churn than the rule needs to produce to do its job. The data in this piece, limited as it is to a six-AMC local sample with a handful of observed cuts and a control benchmark that supports the mechanism without proving it, suggests the churn is not imaginary. It also suggests the fix is not complicated.
This is the mundane version of how markets get better, which is most of the version there is. Most of the improvements to how securities regulation works are not dramatic. They are adjustments to where a line falls, what happens when something crosses it, and whether the rule treats the underlying thing as neat when it is, in fact, continuous. Those are questions about design. The people who drew the 2017 rule were mostly right. They did the larger share of the work. What is left is a smaller share, on the seam, waiting for someone to decide that the right answer to a noisy measurement is not to react to it twice a year as if it were not noisy.