MarketSmith's stock page carries two tabs this service has never read — Deals (exchange-reported bulk and block trades) and Analyst Rating History (every brokerage call with its date, price and target). Both turned out to be reachable without a login. This is what they say about the 132 trades the book actually took.
The intuition — "if a fund is accumulating, that is a tell" — is sound. The data does not carry it. A bulk or block deal is only reported when somebody crosses a threshold in a single trade, and on our universe the names doing that are almost entirely proprietary and high-frequency desks, most of them printing a BUY and a SELL of identical size on the same day. That is market-making, not a view.
| counterparty | deals | what they are |
|---|---|---|
| GRAVITON RESEARCH CAPITAL LLP | 742 | prop / HFT |
| HRTI PRIVATE LIMITED | 350 | prop / HFT |
| QE SECURITIES LLP | 134 | — |
| JUNOMONETA FINSOL PRIVATE LIMITED | 124 | — |
| NK SECURITIES RESEARCH PRIVATE LIMITED | 118 | — |
| XTX MARKETS LLP | 78 | prop / HFT |
| MICROCURVES TRADING PRIVATE LIMITED | 52 | — |
| MATHISYS ADVISORS LLP | 50 | — |
| IRAGE BROKING SERVICES LLP | 38 | — |
| JUMP TRADING FINANCIAL INDIA PRIVATE LIMITED | 38 | — |
| ALPHAGREP SECURITIES PRIVATE LIMITED | 34 | — |
| MUSIGMA SECURITIES | 32 | — |
| SOCIETE GENERALE | 26 | long money |
| SETU SECURITIES PVT LTD | 26 | — |
Long-money institutions appear in the 90 days before an entry on 4 of 132 trades, and in the 30 days before on 0. A predictor absent on 97% of the book cannot be tested, however well it does on the remainder. Real fund accumulation happens below the bulk-deal threshold, over weeks — it shows up in shareholding patterns, which is a quarterly disclosure, not a daily one.
| signal, known before entry | n yes | n no | mean yes | mean no | Δ | p | reading |
|---|---|---|---|---|---|---|---|
| a bullish analyst call, 30d before | 17 | 115 | +5.77% | +1.26% | +4.52 | 0.086 | the strongest single reading — and it does not survive the split below |
| a bullish analyst call, 90d before | 30 | 102 | +3.86% | +1.24% | +2.62 | 0.232 | weaker over a longer window |
| analyst target >15% above price, 90d | 18 | 114 | +4.30% | +1.45% | +2.85 | 0.282 | brokers are optimistic on everything; the threshold adds little |
| 3+ standing bullish ratings | 8 | 124 | +8.88% | +1.38% | +7.50 | 0.049 | n=8, and the win rate is lower — two big winners carry the mean |
| covered by any analyst at all | 67 | 65 | +3.63% | -0.01% | +3.63 | 0.039 | looks like the finding of the study; the split kills it |
| institutional net BUY, 90d before | 1 | 131 | too few to test | present on 1 trade of 132 | |||
Two readings clear p < 0.05 on the whole window. Neither survives contact with the rule this book already learned the hard way: a cross-section is not evidence until it survives a within-period split.
| signal | half | window | n yes | n no | Δ | p |
|---|---|---|---|---|---|---|
| a bullish analyst call, 30d before | first half | 2026-04-01 – 2026-05-07 | 8 | 58 | +0.99pt | 0.772 |
| covered by any analyst at all | first half | 2026-04-01 – 2026-05-07 | 31 | 35 | -0.41pt | 0.845 |
| a bullish analyst call, 30d before | second half | 2026-05-07 – 2026-09-10 | 9 | 57 | +7.65pt | 0.076 |
| covered by any analyst at all | second half | 2026-05-07 – 2026-09-10 | 36 | 30 | +7.60pt | 0.003 |
"Covered by any analyst" reads +3.63pt, p=0.039 across the whole window. Split it: the first half is −0.41pt (p=0.85) and the second is +7.60pt (p=0.004). A signal that is absent in one half and enormous in the other is a description of a period, not of a rule. Removing the three largest winners takes the whole-window gap from +3.63pt to +1.59pt, which says the same thing a different way.
It is also not simply a size proxy: the gap persists across all three liquidity terciles (+5.2, +4.0, +3.5 points). That makes it more interesting and no more tradeable.
Neither deals nor ratings earn a gate. The standing rule applies: rank and record first, earn the gate later.
The endpoint returns every brokerage call with its date, the price at recommendation and the target — richer and better dated than the coverage source the report uses now. Useful as context beside a setup, not as a filter.
On this universe the tab is an HFT print log. Revisit only if the book moves into names large enough for funds to cross in single trades.
Every signal is measured strictly before the entry date of a trade the book actually took, then compared against the realised return of that trade. Significance is a sign-flip permutation test over 5,000 shuffles; the split is the calendar median of the 132 entries. This is a test on fills, not on the candidate population — a fills-only test can only ask "among the trades we took, did this separate them", which is the right question here because both signals were being proposed as a filter on names already selected.