My model stepped out of India. Here’s my best guess why.
The model doesn’t read the news. It noticed the market was unwell and left. Working out what it was reacting to is my job.
When I started the India log in May, I split the country into two stories. Oil was the loud one that I expected to pass. AI was the quiet one that would decide the decade. Both ran through the same place: the household wallet.
I still think that’s roughly the right map. But on 17 September my own India model stepped out of the market, and I owe you two things. An honest account of how much that actually helped. And my best attempt at explaining what it was reacting to, because the model itself can’t tell me.
The scoreboardHow much did it actually dodge?
Less than I’d like, if I’m being straight with you.
The model moved to cash with the Nifty at 23,271. Yesterday the index touched 22,232, and today it’s around 22,520. So the market has fallen about 3% since the model left, and was down 4.5% at the worst point. The model’s book is down 1.6% over the same stretch. Roughly half of the recent slide, avoided.
The bigger picture is humbler. The Nifty peaked at 26,329 on 2 January, so from top to yesterday’s low it fell about 15.6%. The model wasn’t out for most of that. It actually bought back in on 6 July, at 24,430, and stayed in while the index slid almost 5% before it got out again. Since the model went live on 1 June, it’s down 3.2% and the Nifty is down 3.7%. Half a percentage point better. I’d rather show you that than round it up into something it isn’t.
There are two reasons it only caught part of the move, and both are deliberate. It waits until the index is at least 5% off its recent high and the warning signs have stayed on for three days in a row, which keeps it from bolting at every bad week but means it always gives up the first few percent. And only the index part of the book (70%) goes to cash. The 30% in individual stocks stays invested, and those fell too.
So this isn’t a story about a model that saw it coming. It noticed the market was unwell a little after the market got unwell. That’s honestly all I’ve ever asked of it.
Under the hoodWhat it was actually looking at.
This is the part that surprised me when I went back through the files. The model only gets four votes: India’s fear index (India VIX), the Nifty’s one-month trend, its three-month trend, and how far the index sits below its 200-day average.
On the day it left, three of the four were flashing, and all three were about trend. The fear index was sitting at 12.3, which is about as ordinary as it gets. So nobody was panicking. The price had just been drifting the wrong way for long enough that the model stopped giving it the benefit of the doubt.
It also keeps an eye on the rupee, oil, copper and the banks, but none of those get a vote. The rupee one had been flashing since before the exit, and today the rupee is weaker than almost every reading in the model’s history. Oil, at around $100 that day, didn’t even register as unusual by the model’s yardstick.
As of today all four votes are flashing, and the fear index has finally started to climb (14.4). The Nifty is about 9% below its August high.
The way I think about it, the model is a thermometer. It can tell me there’s a fever, but not where it’s coming from. That part is on me, and the rest of this note is my attempt, using the map from the India log. I could be wrong about some of it.
Suspect oneOil: the shock I said would pass.
In Box 1 of the log I called oil the temporary shock: “scary, but it passes.” For a while it looked like I’d got that right.
The war with Iran started on 28 February. Brent went from about $61 at New Year to $118 by the end of March. Then, by late June, the Strait of Hormuz was opening up and Brent was back near $72, below its pre-war level. It’s worth noticing that the model bought back in on 6 July, right in that calm patch. It doesn’t watch oil, but calmer oil showed up as calmer Indian stocks, and that’s what it reacted to.
Then September undid it. The US and Iran started hitting each other’s tankers again in the first week of the month. Brent was back above $100 by the 9th and reached about $109 on the 15th. The model left two days later.
If anything, Brent understates what India is paying. India buys mostly Gulf crude, and those barrels now carry an extra premium because of where they have to sail. India’s own import basket went from about $99 on 2 September to about $124 on 18 September.
The figure that stays with me: from April to August, India bought slightly less crude than a year earlier (0.4% less) and paid 48% more for it, $74.8 billion. Same oil, a much bigger bill.
To be fair to Box 1, the damage so far is smaller than the headlines suggest. India’s current account deficit for April to June was 0.5% of GDP, well below the ~2.2% for the full year that forecasters were pencilling in back in May, though that quarter ended before September’s spike. So “it passes” may still turn out right. It’s just taking a lot longer, and costing more on the way, than I thought it would.
Suspect twoThe rupee: a route I left off the map.
The log so far followed the money through households. What I didn’t draw was the route through the market itself, and I think that’s where most of this year’s damage has been.
Oil is paid for in dollars, so a bigger oil bill means more rupees being sold for dollars. On New Year’s Day a dollar cost ₹89.96. Today it’s about ₹96.72. The rupee is roughly 7% weaker and has spent most of the summer close to its record low.
That matters a lot to foreign investors. The Nifty is down about 14% this year in rupees, but for someone measuring in dollars it’s closer to 20%. And they can get over 5% on a US 10-year Treasury without taking any of that risk.
It’s not hard to see why many of them left. Foreign investors had pulled more than $20 billion out of Indian stocks by the end of April, already more than all of 2025, which was itself a record. They came back for a while over the summer, then started selling again in September.
It also feeds on itself. When foreign funds sell, they convert rupees back to dollars, which pushes the rupee down further, which makes holding Indian stocks look worse for the funds that are still there.
On Wednesday the Reserve Bank of India raised rates by a quarter point, to 5.50%. It’s the first hike since February 2023, and the RBI hinted there may be more. That should help the rupee. But it also lands on the same consumer Box 3 has been tracking, through higher loan payments, at a time when that consumer was already under pressure.
Suspect threeIT: Box 2’s old engine, now in the share price.
In a normal year, IT would have cushioned some of this. Indian IT companies earn in dollars and pay salaries in rupees, so a weaker rupee usually helps them. I half expected IT stocks to hold up.
They didn’t. The Nifty IT index is down about 25% this year, and fell 35% from its December peak to its July low. By late September, TCS was down 35% for the year and Infosys 37%.
One thing the chart showed me that I hadn’t noticed: about four-fifths of IT’s fall this year happened in February, before the war started and before most of this year’s visa news. That doesn’t prove the AI explanation, but it does make me take it more seriously.
Washington has still been the obvious headline:
Introduced in September 2025. A federal judge in Massachusetts struck it down in June. The White House extended it anyway on 18 September, and a second judge, in California, blocked it again on 30 September. A separate proposal for a $103,265 fee is still pending.
On 8 October the US suspended eight companies from PERM, the first step in sponsoring a worker for a green card. TCS, Infosys, Wipro, HCL and Cognizant are on the list.
A proposed 25% tax on US companies’ payments for offshore services. On paper, the most serious of the three.
I’m less sure these explain the selloff as much as the headlines imply. The HIRE Act was introduced in September 2025 and hasn’t had a hearing since; as of April it had no co-sponsors. Infosys filed no PERM applications in 2024 or 2025, and TCS says it doesn’t expect a material impact. And by the White House’s own count, H-1B applications from the big outsourcers had already fallen 92%. These firms have been hiring locally for years.
My honest guess is that Washington is the noise on top of the thing Box 2 was about: the old engine of routine outsourcing being worn down by AI. Back in June that was a thesis about headcount and guidance. Since then the market seems to have started pricing it in. I can’t prove that from the share price alone, so I’ll be reading this month’s IT results for what clients say about next year’s budgets.
Putting it togetherWhere this leaves the log.
I don’t think the model saw three separate problems. I think it saw one chain, from the outside:
Oil goes up, so India needs more dollars.
The rupee weakens, so foreign investors lose money in dollar terms.
They sell, and the rupee weakens further.
The RBI raises rates to steady the currency, which lands on the same households Box 3 is tracking.
IT, which would normally benefit from a weak rupee, is caught up in its own AI story (Box 2), with Washington adding to the worry.
The demand tracker in Box 3 points the same way. Its last reading, on 1 September, was “softening” (−0.75), with tech-corridor housing the weakest marker and oil flagged as rising. That was before September’s oil spike and before the rate hike, so I’d be surprised if the next reading is better.
As for when the model comes back: not because oil falls or the RBI says something reassuring. It needs the Nifty back above its 200-day average (around 24,080 today), its short-term trends pointing up again, and the index within 1% of its recent high. In other words, the market has to recover first, which means it’ll be late getting back in too. It was late in July, and that cost it.
What I’ll be watching in the meantime, by hand:
Oil is where the chain starts.
A sign foreign money has stopped heading for the exit.
Not just a good week.
In this month’s results. That’s the Box 2 question, and it matters more than the visa headlines.
The receiptWhat’s real, what’s noise, and what we still don’t know.
What’s real
- Same oil, a 48% bigger bill (April to August).
- Rupee about 7% weaker this year, near its record low.
- Over $20B of foreign selling by April, already a record.
- The RBI’s first rate hike since 2023.
What’s noise
- The HIRE Act: no hearing, no co-sponsors.
- The PERM suspension: Infosys filed none in two years.
- The H-1B fee: blocked twice in court.
What we still don’t know
- Whether AI is lowering what clients pay for IT work (Box 2).
- How long Hormuz stays contested.
- Whether the RBI’s first hike is also its last.
- Whether “oil passes” (Box 1) still holds.
It noticed. The explaining is mine.
The model avoided about half of the latest leg, not the whole fall. It left because the trend broke, not because it understood oil, the rupee or Washington. The explanation above is mine, and I’ll update this note if the next few weeks show I’ve got it wrong.
Coming up next in the logTwo questions I can’t answer yet.
India has very little AI hardware in its index, which is exactly what has carried the US market. If AI spending ever cools, is India where the money goes next, or is its IT sector too tied to the same story to be a real hedge?
If the Strait of Hormuz properly reopens, the whole chain above runs in reverse: cheaper oil, a steadier rupee, foreign money coming back. How much of that could a rebound recover, how quickly, and would the model even catch it in time, given how late it is by design?