Large, Mid or Small Caps? 18 Years of Indian Data
How much to put in large, mid and small caps, whether P/E or dividend yield can time it, and what cuts the drawdown. Backtests after Indian costs and tax.
How much of an Indian equity portfolio should go into large, mid and small caps? And can valuation (P/E, P/B, dividend yield) tell you when to shift between them? I backtested both, after costs and Indian tax, from 2008 to today. The split mattered a lot. Valuation timing didn’t help. And the cheapest way to soften a crash was gold, not cash.
This follows my ETF rotation backtest, which ended with “just hold”. Holding is the easy part. What to hold is the question this post answers.
The setup
- Three sleeves: Nifty 50, a midcap index and Smallcap 250, all with dividends added back. Index funds, not stock picks.
- A monthly SIP of the same amount, never sold. That’s how most people actually invest.
- After costs and tax: 0.1% per trade, Indian capital gains tax, paid every financial year.
- Two periods plus rolling windows: 2012-2019, 2020-2026, the full run from 2008, and every 5- and 10-year SIP window by start year.
Do active funds beat the index?
Before choosing a split, I checked whether active funds are worth it in each category. I compared about 140 direct-plan funds with their index, weekly, since 2013.
- Small caps: often yes. The better small-cap funds beat Smallcap 250 by 2-6 points a year over 10 years, and did so in 80-100% of rolling 5-year windows.
- Mid caps: rarely. A few funds led by 2-4 points. The median fund roughly matched the index, and the worst trailed it by 3-4 points for a decade.
- Flexi and focused: mixed. A few funds were consistent. Most were close to Nifty 500.
Two warnings. Closed and merged funds are missing from the data, so the field looks better than it was. And last-5-year leaders change with style cycles. I gave more weight to 10-year records and rolling win rates.
The split drives the result
Same monthly SIP into each mix of large / mid / small:
| Split L/M/S | 2012-19 | 2020-26 | 2008-now | Worst drop | Median 10y | Worst 10y |
|---|---|---|---|---|---|---|
| 100/0/0 | 10.7% | 9.9% | 10.8% | -60% | 12.3% | 9.9% |
| 50/25/25 | 9.9% | 13.3% | 12.6% | -67% | 14.1% | 9.2% |
| 35/35/30 | 9.7% | 14.3% | 13.1% | -70% | 14.9% | 9.0% |
| 30/30/40 | 9.4% | 14.6% | 13.2% | -71% | 15.5% | 8.8% |
| 20/40/40 | 9.3% | 15.2% | 13.6% | -72% | 15.9% | 8.8% |
| 0/50/50 | 9.0% | 16.4% | 14.1% | -75% | 17.0% | 8.4% |
Yearly returns (XIRR) after costs and tax.
Over 18 years, each step toward mid and small caps added about half a point a year, and a deeper crash. Every mix with mid and small caps beat Nifty alone in 8 of 9 ten-year windows.
But 2012-2019 alone ranks them the other way round. If you’d judged by those eight years, you’d have picked Nifty only. One period can mislead you in either direction.
The middle of the table is flat. 35/35/30, 30/30/40 and 20/40/40 are within half a point of each other. That gap is smaller than the noise in 18 years of data, so the exact number matters less than picking one and keeping it.
Can valuation choose the split?
Mid caps get expensive, then small caps, then large caps. So the natural idea is to put new money into whichever sleeve is cheapest. NSE publishes P/E, P/B and dividend yield for each index since 2013. I tried two versions:
- Each sleeve against its own history. Cheap mid caps get more of the SIP.
- Mid and small against large. If mid-cap P/E is high relative to Nifty’s, shift toward large.
I tested each with new money only (no selling), and with a yearly rebalance of the whole portfolio (which pays tax).
| Signal | New money only | With yearly rebalance |
|---|---|---|
| Own history (P/E, P/B, yield) | -0.8 to +0.1 | |
| P/E vs Nifty | 0 / 0 | -0.4 / +1.0 |
| P/B vs Nifty | -0.1 / -0.6 | +0.4 / -0.9 |
| Dividend yield vs Nifty | 0 / -0.1 | +0.6 / -0.5 |
Points a year against the fixed 35/35/30 SIP, for 2012-19 / 2020-26.
The tilts were not timid: small caps got anywhere from 0% to 91% of a month’s SIP. Still nothing held up. With new money only, the result barely moved. With rebalancing it moved, but the sign flipped between periods. The best signal before 2020 lost after, and the best after 2020 lost before. That’s noise, and it pays tax.
The measures also disagree today. By P/E, small caps look cheap. By dividend yield, large caps do. Small-cap P/E is noisy because loss-making companies distort it, and NSE switched to consolidated earnings in 2021, which lowered every index’s P/E overnight.
”Hold cash when it’s expensive” halves the drop, not the return
This one surprised me, because it feels obviously right. In the earlier experiments I held Nifty at 100%, 80% or 50% depending on its P/E, with the rest in a liquid fund:
| 2012-19 | 2020-26 | Worst drop 2020-26 | |
|---|---|---|---|
| Nifty, always invested | 12.3% | 11.1% | -36% |
| Nifty by P/E band, rest in liquid | 11.7% | 11.8% | -18% |
It halved the worst drop. It didn’t add return. Why:
- Expensive markets stay expensive for years. Being half out during 2021 cost 16 points that year.
- The dip you wait for often stays above where you got out. The market rises 30%, falls 15%, and you buy back higher than you sold.
- Crashes recover fast. 2020 was over in months. After that, the cash sits at 6%.
Parking a monthly SIP in liquid when the market looked expensive, and deploying on cheapness, did worse than investing every month in all 8 settings I tried.
When were the worst drops?
Taking the 30/30/40 mix as one lump sum from 2008, these are the falls of more than 20%:
| Peak | Bottom | Drop | Back at the peak |
|---|---|---|---|
| Jan 2008 | Mar 2009 | -70% | May 2014 |
| Aug 2015 | Feb 2016 | -21% | Jul 2016 |
| Jan 2018 | Mar 2020 | -43% | Dec 2020 |
| Jan 2022 | Jun 2022 | -21% | Jun 2023 |
| Sep 2024 | Mar 2025 | -21% | Aug 2026 |
Nifty alone fell 59% in 2008 and was back by October 2010. Mid and small caps fall harder and take years longer to recover. 2018-2020 was a mid and small-cap bear market that Nifty barely noticed.
What softens the drop
I added three overlays to the 30/30/40 mix:
| Overlay | Lump sum 2012-19 | Lump sum 2020-26 | Worst drop 12-19 / 20-26 | SIP 2020-26 |
|---|---|---|---|---|
| None | 12.8% | 17.0% | -28% / -38% | 14.6% |
| +15% gold, rebalanced yearly | 12.4% | 17.5% | -23% / -32% | 15.7% |
| +25% gold, rebalanced yearly | 11.7% | 17.7% | -21% / -28% | 16.3% |
| Sleeve below its 200-day average goes to cash | 7.3% | 12.7% | -22% / -24% | 9.5% |
- Gold cut the worst drop by 5-10 points at about the same return. It usually rises when Indian stocks crash. It isn’t free money: it cost a little before 2020 and helped after, partly because of gold’s own big run. Think of it as insurance.
- A trend filter cut 2008 from -70% to -42%, but cost 4-5 points a year. Too expensive.
- Not selling matters more than either. A SIP that kept running through 2008 recovered far faster than the lump sum.
What I’m doing with it
- A fixed split of 30/30/40. Rebalanced with new money, and by selling only when a sleeve drifts more than 10 points.
- About 10% in gold, rebalanced once a year.
- Few funds: active only in small caps, where the edge held up. Index funds or a couple of long-record funds elsewhere.
- No valuation timing. I checked P/E, P/B and dividend yield four different ways. None of them earned its tax.
Check your own mix
- Test the split over at least one full mid and small-cap cycle (2008-2014, 2018-2020), not just the last five years.
- Compare each active fund with its own index over 10 years and rolling 5-year windows.
- Before adding a timing rule, see whether it beats the same fixed split after tax in both periods.
- Decide in advance what you’ll do in a -40% year. Stopping the SIP then costs more than any split choice.
Takeaways
- The large/mid/small split moves long-term returns more than any timing rule I tested.
- P/E, P/B and dividend yield didn’t reliably choose the split, and holding cash when expensive bought a calmer ride, not more return.
- Gold was the cheapest way to cut the drawdown.
If you’ve found a valuation rule that beats a fixed split in both periods after tax, send it to me and I’ll run it through the same engine.
This is a personal research note, not investment advice. Backtests describe the past; they don’t promise anything. Index data: NSE and Yahoo Finance; fund NAVs: mfapi.in.