The factors doing the actual stock-picking

This piece is a follow-up to the multi-screen quant shortlist published on this blog on 24 July 2026. That shortlist, and the model portfolios built from it, come out of a rules-based screening approach that stacks eight distinct factor methodologies rather than relying on any single one: Darvas-box breakouts (price clearing a defined consolidation range on rising volume), Piotroski quality scoring extended with a debt-reduction check (profitability, leverage, and efficiency trends, not just a snapshot), Coffee Can investing (long-holding-period quality compounders, named for the old habit of putting valuables in a tin and forgetting about them), the Magic Formula (ranking on the combination of earnings yield and return on capital), a momentum/golden-cross overlay (moving-average crossovers confirming an established trend rather than a one-day spike), breakout-quality grading (an A-through-F score on how clean a technical setup is), a machine-learning bullish-flag classifier clipped to reduce false positives, and a hard liquidity gate that excludes anything too thin to trade at size. Positions that clear enough of these filters simultaneously get grouped by sector and market into co-movement "model portfolios," sized by inverse volatility and capped at 25% per name so no single holding can dominate a bundle. The question this piece asks of that stack, across the India and US markets: does combining eight factors this way actually produce a portfolio you couldn't get from a plain index fund, and if so, is it actually winning?

The headline track record

Across 125 live picks generated by this factor stack, 94 are currently showing a gain and 31 a loss against their entry price — a 75% hit rate — with an average return of +9.7% since each name was flagged, over a median holding period of roughly 38 days. That is a genuinely strong top-line number for a rules-based approach, and it's the number that would normally get quoted on its own. It needs a qualifier, though: a high hit rate on individual picks is consistent with a rising market lifting most positions regardless of stock selection. The more diagnostic question — whether combining these factors at the portfolio level is adding anything a plain index fund wouldn't — is answered by the alpha figures below, not by the hit rate.

Seven bundles, one clear outperformer

The system's seven model portfolios were all formed on 23 July 2026 using inverse-volatility weighting capped at 25% per name. Ranked by alpha — return in excess of the broad market over the same window, not just raw return — only one bundle is generating real alpha:

Six of seven bundles are producing a positive alpha sign only when you don't adjust for the market's own move — four of the six are still up in absolute terms, but a rising broad market has been carrying them faster than they've been carrying themselves. Only the five-name US Healthcare bundle (BAX plus four others) is compounding faster than the market it's implicitly being measured against. That's one clean success out of seven bets on a construction methodology, which is a smaller number than the 75% headline hit rate on individual picks would suggest, and it's the kind of gap that only becomes visible once return and alpha are pulled apart.

Differentiated, but not for the reason that matters yet

Checking each model portfolio's constituent overlap against the closest public benchmark gives an unambiguous, and on its own terms reassuring, finding: every US bundle runs at 100% active share against its closest sector SPDR. The Healthcare bundle shares just 1 of 5 names with XLV's 63 holdings; the Financial Services bundle shares 0 of 5 with XLF's 80; Technology+Industrials shares 1 of 6 with XLK's 77; Energy shares 0 of 6 with XLE's 24. The India Financial Services + Consumer Cyclical bundle overlaps just 1 of 10 names with Nifty Financial Services, though 8 of 10 sit inside the broader Nifty 500 — differentiated on sector composition, but still a large/mid-cap book rather than a small-cap discovery engine. None of this is closet indexing; the factor stack is finding names the passive funds simply don't hold.

But the alpha table in Section 3 shows that differentiation from a benchmark and outperformance of it are two separate claims, and constituent overlap only answers the first one. A portfolio can be built from entirely different names than an index fund and still track that index's direction and magnitude fairly closely, which is roughly what's happening in five of the six alpha-negative bundles above — genuinely distinct holdings, moving in a direction the broad market would have handed you anyway.

Where today's move is actually concentrated

Sector dispersion in the current session is wide. Technology is the clear leader (median +3.10% on the day, +5.53% over five sessions, 69% of the sleeve green), and it's home to the single best individual mover across both books — BULL, up 9.0% today and +17.4% since it was first flagged. Energy is the second-hottest sector today but its five-day median is actually negative (-1.63%), which a one-day snapshot alone would miss entirely: today's bounce sits inside a weaker short-term trend, not a resumption of one. Communication Services is the weakest cohort in the current session, flat-to-down across the board with zero green names.

That dispersion isn't evenly distributed by liquidity, either. The most liquid tier (T1, roughly $12M+ average daily value) is essentially flat today at a 0.00% median move, while the lower-liquidity T3 tier (down to roughly $600k/day) is where both today's best mover and its worst single-stock outcome actually sit. The volatility in this session is a lower-liquidity-tier story, which is exactly where slippage risk is already structurally higher — worth flagging on its own, independent of any individual name.

What the momentum-decay exit rule catches, and the one case it can't distinguish

Alongside the entry-side factor stack, the same system runs a momentum-decay exit rule: a position that has broken down from its own trend gets flagged to sell, regardless of whether it's sitting on a gain or a loss. Eighteen of today's twenty exit signals are closing out positions still showing a gain since they were flagged — profit-taking working exactly as that rule is designed to, not a warning sign. One name, though, doesn't fit that description: one US small-cap position is down roughly 87% today alone and -73.4% since it was first flagged, by a wide margin the worst outcome in the current book. That's the one structural blind spot in a symmetric exit rule — it correctly flags both "take profit" and "cut the loss" as the same action, "sell," which is the right mechanical response but means a single glance at "18 sells today" understates that seventeen of them are wins being banked and one is a loss being contained.

Is a fixed sell-zone eviction window cutting winners too early?

Beyond the daily entry/exit signals, the system also runs a longer eviction rule: a position gets dropped from active tracking after six consecutive sessions inside its sell zone. 74 evictions have accumulated under that rule to date. Most evicted names continue to underperform after being dropped, which is the rule working as intended — persistent sell-zone behaviour is, on average, informative. But at least four recent evictions bounced meaningfully afterward, recovering between roughly 10% and 21% once dropped. Four apparent false positives against roughly thirty evictions with visible post-eviction data isn't damning on its own, but it's a legitimate, testable question for this kind of rule generally: whether a fixed multi-session threshold is sometimes cutting names that were about to turn, rather than names that were genuinely deteriorating. This piece doesn't have enough evicted-name history to answer that either way — it's raised as an open methodological question, not a conclusion.

What would fix this

• SEBI should require registered research analysts and investment advisers who publish model portfolios to report returns alongside a named benchmark's return over the same window, so a high hit rate on individual picks, like the 75% here, is never presented without the alpha figure that can tell a different story.
• NSE Indices should publish full daily constituent lists with weights for every index it maintains, free and in machine-readable form, so active share against Indian benchmarks can be calculated properly rather than from membership lists alone.
• AMFI should publish a consolidated, machine-readable file of the monthly portfolio holdings of every Indian index fund and ETF, so overlap with Indian sector and broad-market funds can be checked as easily as this piece checked US bundles against SPDR daily holdings.

These are proposals from this blog, not recommendations any agency has adopted.

Sources and caveats