Drift-band vs monthly rebalancing with Indian costs and capital-gains tax
Short answer: A 5-point drift band beat monthly rebalancing in most windows, by 0.1–0.6 percentage points a year.
Published 2026-10-30 · original backtest on Indian market data by TRINETRA
Date: 2026-10-30 · Script: engine/scripts/bench_rebalance_band.py Outputs: data/research/BENCH_rebalance_band_1000000.json, …_10000000.json Idea source: trader-skill trader-rebalance (drift band, sells before buys, min order).
Setup
- Mix NIFTY 60 / Gold 25 / NASDAQ 15, daily closes 2007-09-17 → 2026-10-27. Gold and NASDAQ are USD series; FX is ignored (it hits every arm the same).
- Costs: delivery STT, stamp, exchange and DP charge from
backtest/costs.py. - Tax: 12.5% LTCG above the s.112A ₹1.25 lakh/FY exemption, average cost basis. This is a lower bound: lots sold inside 12 months pay 20% STCG, which the model does not split out. Rates checked 2026-10-30 against current secondary sources (no change since 23 July 2024).
- 29 rolling 5-year windows, one every 6 months. They overlap heavily, so they are about 4 independent periods. Treat window win-rates as indicative, not as significance.
Result
| arm | ₹10L full CAGR | ₹10L tax (% of start) | ₹1Cr full CAGR | ₹1Cr tax | median 5y max DD | beats monthly (₹10L / ₹1Cr) |
|---|---|---|---|---|---|---|
| never | 9.82% | 0% | 9.82% | 0% | −17.8% | 31% / 52% |
| monthly | 10.32% | 14.9% | 9.97% | 30.6% | −15.1% | — |
| quarterly | 10.34% | 10.3% | 10.10% | 22.8% | −15.4% | 48% / 55% |
| annual | 10.53% | 3.9% | 10.36% | 13.4% | −15.4% | 48% / 69% |
| band 2pt | 10.37% | 12.9% | 10.08% | 26.5% | −15.3% | 62% / 76% |
| band 5pt | 10.65% | 8.0% | 10.44% | 18.5% | −15.5% | 83% / 93% |
| band 10pt | 10.41% | 7.4% | 10.31% | 12.5% | −16.3% | 72% / 86% |
What it says
- Rebalancing is a risk tool, not a return tool. Any rebalancing rule cut the median 5-year drawdown by about 2.5 points versus never rebalancing, at about the same return.
- The method matters little: 0.1–0.6 points of CAGR. A 5-point band is the most consistent winner over monthly, but nobody should sell it as an edge.
- In India, tax is the real cost of rebalancing, and it scales with portfolio size. STT and stamp duty are rounding errors next to it. The ₹1.25 lakh exemption absorbs most of an annual or banded rebalance for a ₹10 lakh book, and does not for a ₹1 crore book.
- Not tested here: tax-gain harvesting. India has no wash-sale rule, so realising up to ₹1.25 lakh of LTCG each FY and rebuying resets the cost basis for free. The most it saves is 12.5% × ₹1.25 lakh ≈ ₹15,600 per person per year. That's arithmetic, not a backtest result.
Product decision
If TRINETRA ever proposes rebalancing trades, use a 5-point absolute drift band, do sells before buys, and show the estimated tax beside each sell, checked against the year's remaining ₹1.25 lakh exemption. A calendar-monthly default would cost a ₹1 crore investor about 31% of starting capital in tax over 19 years, for a drawdown benefit they also get from the band.