In 2022, I bought HBL Engineering at around ₹90–100.
Not because I had a thesis. Because I had an interest. Defence electronics, railways, niche components — it seemed like a good business to own in a cycle that was clearly building. So it joined the portfolio at 1–2%, like everything else did at the time.
HBL went from ₹100 to ₹500 over the next two years. My portfolio barely noticed.
The stock delivered. The portfolio didn't capture it. That gap — between a call being right and a portfolio actually benefiting from it — is where most of the self-inflicted damage in direct equity investing lives. Not in the bad calls. In the right calls that were sized like guesses.
A 1–2% position is the honest size of an interest. It says: I think something good might happen here, but I haven't pushed this idea hard enough to know exactly what or why. When that position doubles or triples, it moves the portfolio by a rounding error. The research time cost you the same whether you owned 1% or 10%. The return is nowhere near the same.
What caused the position to be undersized — and what changed?
The inflection came when HBL was around ₹500. By then I had made a decision that had nothing to do with HBL specifically: the 80-stock portfolio had to go. I collapsed it to eight positions. Each stock that survived had to clear a real threshold — a specific thesis I could state without notes, KPIs I could name that derived from that thesis, and a written exit condition that wasn't a price level.
HBL survived that process. Not because it had already run 5x — a stock's past performance is not a thesis. It survived because when I finally did the proper research, I found a genuinely good business.
Defence electronics with long qualification cycles creating switching costs that commodity competitors can't easily bridge. Railway component supply to a customer base expanding aggressively through a government capex programme with visible multi-year runway. A management team that had been building quietly for years before the cycle caught up with them. The kind of business where the moat isn't a single factor — it's the accumulated complexity of being deeply embedded in critical infrastructure.
The thesis I could write: HBL's revenue will compound as defence and railway electronics order inflows grow through the ongoing government capex and indigenisation cycle, and the long qualification lead times for its niche components protect margins from commodity competition.
The KPIs I could name: order book additions and executable order backlog each quarter, execution margins on electronics orders, and revenue mix shift from legacy battery business toward electronics.
The exit condition I could write: if electronics order intake guidance reverses for two consecutive quarters, or if a structural policy shift reduces the defence indigenisation mandate, the load-bearing assumption has changed.
All of that was available at ₹100. The business was the same business. The information was there. The work I did at ₹500 could have been done in 2022. I just hadn't demanded it of myself.
What does position size reveal about actual conviction?
This is what position size as a conviction statement actually means.
It doesn't mean you should always size aggressively. It means that the size of a position is an honest record of how much you know about a business — whether you treat it that way or not. A 1–2% position in a business you've genuinely interrogated is a deliberate tracking position. A 1–2% position in a business you haven't is an interest that happens to be in your portfolio.
They look identical in the holdings table. They produce completely different outcomes when the thesis plays out.
One caveat worth naming: position size as a conviction signal is most reliable for investors who have a deliberate, consistent framework for sizing. For an investor still developing that framework, size alone doesn't always tell the whole story. But for anyone who has thought seriously about how to build a portfolio, the gap between what you know and what you're staking on it tends to be the most honest diagnostic available.
What was different about the question asked at ₹100 vs ₹500?
The harder part to sit with is what didn't happen between 2022 and 2024.
It wasn't that the information about HBL was unavailable. The annual reports existed. The concalls were public. The order book data was disclosed every quarter. The indigenisation push in defence was clearly directional. None of what I found at ₹500 required access to anything that wasn't there at ₹100.
What changed was the question I was asking. At ₹100, the question was: does this business seem good? At ₹500 — forced by the decision to collapse the portfolio — the question was: what specifically has to stay true for this business to keep compounding, and what would tell me if it wasn't?
The second question produces a thesis. The first produces a position at 1–2%.
Most investors spend a lot of time trying to find better stocks. The more productive question is usually: am I asking the right things about the stocks I already own? The information is often available. The interrogation is what's missing.
What does the HBL experience mean for how to approach new positions?
The HBL position I hold now is sized meaningfully. It survived the cut not because of what it had already done, but because of what the research confirmed it still was. It continues to compound, and the portfolio captures it now in a way it couldn't in 2022.
But the move from ₹100 to ₹500 — the part I was right about from the beginning — is gone. The interest was correct. The conviction arrived too late to use it.
The pattern isn't unique to this stock. Any portfolio that has accumulated positions through interest rather than interrogation has the same structure — a collection of maybes, sized like maybes, that will deliver maybe-level returns even when the calls are right. The fix isn't finding better stocks. It's interrogating the ones you already own with the same rigour you'd apply if you were forced to cut the portfolio in half.
Because that's exactly what the four questions are designed to do.