Blending Growth Fundamentals With Momentum on Nifty 500: Tilting the Blend Cost 0.79 Points

Lineage: Strategy 24 opens what 21 to 23 left hardcoded: term weights, and order versus magnitude.

The premise

Three earlier grids blended growth metrics into one cross-sectional score and averaged the terms equally. None tested whether equal was right. The ceiling had walked 41.40% to 41.41% to 46.46% across those three, each time by opening an axis the previous grid had pinned, so weighting was the obvious thing to attack next.

Two questions follow. Does promoting one growth term above the others help, and is a rank or a z-score the better way to combine them? A rank keeps the ordering; a z-score keeps the distance too.

How the strategy works

Every company is scored on a blend of growth terms, each measured over one to four reported quarters against the same quarters a year earlier, drawn from revenue growth, profit growth, margin change and operating leverage. Each term becomes a cross-sectional percentile rank, or a z-score clipped at four standard deviations, then is weighted, summed and divided once by the weight total. A company unscoreable on any term is dropped, so three-term blends pick from a thinner field and earn 21.06% at the median against 23.08%.

The score decides which companies are eligible, price decides when they are bought. Two or three times the target size is taken from the top of the ranking, filtered through a moving-average gate with independent entry and exit rules, and the survivors re-ranked on volatility-adjusted momentum over 63, 126 or 252 days. The best five, ten or twenty are held. A 20% trailing stop closes a position and its proceeds go to the survivors in proportion to each one's gain. Rebalancing fires on the quarterly earnings refresh.

The setup

25,920 simulations over 1,291 Nifty 500 names, 1 January 2016 to 8 May 2026, crossing 180 weighted specifications with three portfolio sizes, three lookbacks, four price gates, two pool widths and two transforms. ₹10 lakh is funded once on day one. From there the book rotates: a name falling out of the target is sold, the gain realised, and 10% tax paid at that point. Costs are 0.03% per side and idle cash earns 5%.

The benchmark is not the Nifty 500 index. It is a costlessly daily-rebalanced equal-weight portfolio of the eligible universe, taxed once at exit, not at every rebalance: 12.04% XIRR, a 57.28% drawdown, 0.21 Calmar, ₹32.44 lakh. No rebalancing cost and deferred tax flatter it, deliberately.

The best combinations

Six rows matter: the three strongest by XIRR and the three by Calmar. All score by rank, five hold only five names, and none returns under 34.78%. The exact parameter string behind each is withheld.

They share concentration and patience. Four use the 252-day lookback, and both Calmar leaders carry equal weights. The strongest row returned 42.08% across the first half of the window and 52.14% across the second, which meant holding five stocks through a 43.99% drawdown.

What the weighting actually does

Tilting is the axis this grid was built to test, and it loses. Across 23,328 matched pairs, each tilted row against the equal-weight row sharing its specification, gate, size and lookback, a tilt costs 0.79 points of XIRR at the median and wins 43.52% of pairs.

Which term is promoted decides how much is lost. A revenue-led tilt is the only one that pays, adding 0.45 points across 9,504 pairs at a 53.54% win rate, while margin-led tilts lose 2.13 points and oplev-led tilts lose 2.40. Revenue is the slowest-moving metric, so promoting it sharpens an ordering already stable, where a margin tilt amplifies a number that reverses sign on one weak quarter.

Magnitude scoring fails the same test. Across 12,960 matched pairs a z-score costs 0.83 points against the rank and wins 43.69% of the time, and it does not raise the ceiling: 46.86% for the best rank row against 42.32% for the best z-score row. Discarding distance is the useful part of a rank, not a limitation.

Year by year

One year dominates the leaderboard, and that belongs beside the headline. The Spearman correlation between a row's 2020 return and its full-window XIRR is 0.560, and 0.691 in the earlier comparable grid. Yet that same 2020 return correlates with the other ten years compounded at -0.048, so winning 2020 said nothing about any other year. 2022 is the more durable evidence: 95.97% against a flat benchmark, in a year when the gate declined to buy names already falling.

Does the ranking survive

It does not, the most important finding here. The correlation between first-half and second-half XIRR is -0.09, so where a combination ranked in the first window says nothing about the second. The top 50 rows picked on the first half returned a median 23.11% in the second, against 23.53% grid-wide. Random selection would have done marginally better. The row leading the first half at 56.98% returned 27.66%.

Where it fails

2025 was structural rather than unlucky. 97.90% of all 25,920 combinations finished that year negative, at a median of -21.90% against the benchmark's -0.26%. A momentum re-rank buys whatever led most recently and a 20% stop sells it into the reversal, so a year that rotates leadership repeatedly charges this design twice. 2018 repeated the pattern at -21.25%.

Concentration is the second failure, and it is a choice, not a flaw. Five-name portfolios hold both the best row in the grid and its worst drawdown of 83.73%, and earn a lower median than ten-name portfolios, 20.17% against 23.03%. The weakest combination lost money outright: -2.35% XIRR on a 72.66% drawdown.

Verdict

The blend of growth fundamentals with price momentum works here, and a margin of 46.86% against 12.04% survives a reasonable amount of scepticism. Neither refinement this grid was built to test does. Tilting cost 0.79 points and worsened the harder it was applied, magnitude scoring cost 0.83, and neither raised the ceiling. Equal weights and plain percentile ranks, used by three earlier grids without ever being tested, are what the evidence endorses. Hold ten names not five, expect a median nearer 23% than 46%, and read 46.86% as a figure that leaned on 2020.

Appendix: assumptions and methodology

• Universe. 1,291 Nifty 500 names, 1 January 2016 to 8 May 2026, scored cross-sectionally each day.
• Benchmark. A costlessly daily-rebalanced equal-weight portfolio of the eligible universe, taxed once at exit, not the index: 12.04% XIRR, 57.28% drawdown, 0.21 Calmar.
• Grid. 180 weighted specifications, three sizes, three lookbacks, four gates, two pool widths, two transforms: 25,920 runs.
• Execution. ₹10 lakh lump sum on day one, so XIRR and CAGR coincide. Costs 0.03% per side, 10% tax per realised gain, idle cash 5%. Halves split at 7 May 2021.
• Metrics. XIRR on dated flows, drawdown peak-to-trough on daily value, Calmar as XIRR over absolute drawdown.
• Held fixed. Momentum family, the 20% stop, earnings-refresh rebalancing and band policy, pinned by earlier grids.
• Not modelled. Slippage beyond 0.03%, market impact at five names, STT, dividends, survivorship, small-cap fills.
• Withheld. The exact parameter string for each row, per-term quarter counts and gate definitions.