Research register

Let the evidence speak in full.

Each question with its measure, its test and its verdict so far.

Every postulate, tested in the open.

1 tested and carried information · 7 tested with no evidence · 5 running · 0 pending

What each status means

Running: measured and published; too early for a verdict
Tested — carried information: tested with a correction for testing many ideas at once, on non-overlapping periods and beside a random control, and the effect held over the sample tested
Tested — no evidence: tested the same way; no reliable effect was found
Pending: designed, not yet measured

Paid tier

This section holds how each postulate was tested — the harness, the controls, the corrections — and the numbers behind each verdict.

Fund postulates

Claims about what a mutual fund’s own record and disclosures show. Each is measured for every fund and printed on the fund page as a description of the past. Each was also tested as a forecast of which fund beats its peers next.

P1 Churn, and what it boughtTrading costs something certain; a fund that trades more should show extra return that pays for it. Tested — no evidence

Measured: Turnover computed from consecutive monthly portfolios: half the sum of every change in a holding’s weight, over a year. Shown beside the excess return per unit of turnover.
Data: Monthly portfolio disclosures from 38 fund houses, daily NAVs (net asset value: the price of one unit) from the scheme registry, and monthly factsheets. Refreshes: Monthly, about ten days after each month ends.
Test: How it was tested, and the numbers: the paid tier.
Verdict so far: Shown as a description. As a forecast it points the wrong way: the lowest-turnover fifth trailed the highest by 1.72 percentage points a year (p = 0.017), which does not survive the correction for testing many measures.
Started: 22 Sep 2026
Why it might matter: Fund houses define their own turnover figures inconsistently; this one is computed the same way for every fund, so two funds can be compared on it.
Why it might not: Month-end snapshots miss everything bought and sold inside a month, and a fund forced to sell by withdrawals is charged for churn it did not choose.
On the site: Funds, by turnover · Does any of it predict?

P2 Conviction or inertiaLong-held top-ten positions mark a portfolio someone chose; a top ten that changes every quarter is a reaction. Tested — no evidence

Measured: The median number of months each current top-ten holding has been in the top ten without a break.
Data: Monthly portfolio disclosures from 38 fund houses, daily NAVs (net asset value: the price of one unit) from the scheme registry, and monthly factsheets. Refreshes: Monthly.
Test: How it was tested, and the numbers: the paid tier.
Verdict so far: Shown as a description. As a forecast: top fifth minus bottom fifth +0.19 percentage points a year, indistinguishable from zero (p = 0.74).
Started: 22 Sep 2026
Why it might matter: Persistence is how long a manager keeps a conviction; it separates a decision from drift.
Why it might not: Persistence without a different portfolio from the index is inertia, and the measure cannot yet tell the two apart.
On the site: Funds, by how long top holdings are kept · Does any of it predict?

P3 Hit rate of the top tenA manager’s largest positions, judged one by one over the next six months, say more than the fund’s blended return. Tested — no evidence

Measured: Of each month’s ten largest holdings, the share that beat the typical stock held in the same category over the following six months, and by how much on average.
Data: Monthly portfolio disclosures from 38 fund houses, daily NAVs (net asset value: the price of one unit) from the scheme registry, and monthly factsheets. Daily exchange closing prices. Refreshes: Monthly.
Test: How it was tested, and the numbers: the paid tier.
Verdict so far: Shown as a description of six months already past. As a forecast: top fifth minus bottom fifth −0.66 percentage points a year (p = 0.64).
Started: 22 Sep 2026
Why it might matter: A fund can beat its category on two positions and lose on eight; the count and the margin show what the blended return hides.
Why it might not: Six months is short, sizing matters more than counting, and a long-held winner is counted again every month it is held.
On the site: Funds, by top-ten hit rate · How often is not how much

P4 Consistency, not a single end dateA fund that beats its category in most rolling three-year stretches has a record; one that beats it in a few has an end date. Tested — no evidence

Measured: Every three-year stretch since the fund had a NAV, one starting each month: the share it beat the category’s typical fund, and the average margin across all of them (win share × average win + loss share × average loss).
Data: Daily NAVs from the scheme registry, and the category each fund is filed under. Refreshes: Monthly, at each month end.
Test: How it was tested, and the numbers: the paid tier.
Verdict so far: Shown as a description. As a forecast: top fifth minus bottom fifth −0.68 percentage points a year on the three-year share (p = 0.56). Winning often is also not winning by much: across 523 funds the share of stretches won correlates only 0.32 with the average margin.
Started: 22 Sep 2026
Why it might matter: Hundreds of overlapping stretches cannot be flattered by one end date the way a headline five-year figure can.
Why it might not: The stretches overlap, so the true number of independent observations is a handful; a fund that changed manager carries a record only partly its own.
On the site: Funds, by average margin · How often is not how much

P5 Cost, measuredThe Regular plan (sold through a distributor, who is paid from the fund every day) and the Direct plan hold the same portfolio; the gap between their NAVs is the cost. Tested — carried information

Measured: The yearly growth of the Direct plan’s NAV minus the Regular plan’s, over the same dates.
Data: Daily NAVs for both plans of every fund, from the scheme registry. Refreshes: Monthly.
Test: How it was tested, and the numbers: the paid tier.
Verdict so far: The one reliable number in the study: choosing Direct over Regular was worth a median 1.21 percentage points a year (middle half 0.91 to 1.58), measured on 33,028 fund-dates — larger than every skill measure combined.
Started: 22 Sep 2026
Why it might matter: It is the one number on a fund page that is knowable today rather than hoped for, and it compounds for as long as the units are held.
Why it might not: A Regular plan pays for advice; if that advice stops a sale in a fall, it can be worth more than the fee. The cost is visible and that saving is not.
On the site: What the Regular plan costs · Funds, by Regular-plan mark-up

P6 Size against edgePast a certain size a fund cannot buy what a small one can, and should charge less in return. Tested — no evidence

Measured: Where the fund sits by size in its category, how much of its growth was new money rather than returns, and its expense ratio (the yearly fee, as a share of assets).
Data: Monthly portfolio disclosures from 38 fund houses, daily NAVs (net asset value: the price of one unit) from the scheme registry, and monthly factsheets. Published quarterly average assets and daily expense ratios. Refreshes: Monthly; assets quarterly.
Test: How it was tested, and the numbers: the paid tier.
Verdict so far: The only measure with a weak signal: smaller funds did slightly better inside their own category (+1.18 percentage points a year; t = 2.07 after controls), but it does not survive the multiple-testing correction and is tangled up with a decade in which small and mid-sized companies did well.
Started: 22 Sep 2026
Why it might matter: A fund that has tripled through new money while charging above its category median has kept the benefit of scale for the fund house.
Why it might not: Scale pays for research teams, and money follows past performance, so a fast-growing fund is usually one that did well. In India the fee cap falls as a fund grows, so fee and size are the same fact seen twice.
On the site: Funds, by expense ratio · Does any of it predict?

P7 Whose record is thisA record earned under different managers is not evidence about the ones running the fund now. Tested — no evidence

Measured: How long each manager named on the monthly factsheet has been named, dated to the month the change was disclosed. Where the archive starts after a manager did, the figure is a floor and is marked so.
Data: Monthly factsheets from 38 fund houses. Refreshes: Monthly.
Test: How it was tested, and the numbers: the paid tier.
Verdict so far: Shown as a description. As a forecast: top fifth minus bottom fifth −0.53 percentage points a year (p = 0.57), on the shorter sample from 2018 where manager names are available.
Started: 22 Sep 2026
Why it might matter: A five-year figure on a fund whose longest-serving manager joined two years ago is the fund house’s record more than the person’s.
Why it might not: Fund houses have processes and research desks that outlast individuals, and a factsheet names the manager of record, not necessarily the person making the calls.
On the site: Managers · Whose record is it?

P8 Manager decisionsScore a fund’s decisions rather than its holdings: adding to a stock that then beat its benchmark, or cutting one that then lagged, is a right call. Tested — no evidence

Measured: For every pair of monthly portfolios, the active change in each weight — the actual weight minus the weight the position would have had with no trading — classed as entry, add, trim or exit, then judged over the next three months against stocks of the same size tier and sector. Changes forced by money arriving or leaving are flagged.
Data: Monthly portfolio disclosures from 38 fund houses, daily NAVs (net asset value: the price of one unit) from the scheme registry, and monthly factsheets. Daily prices adjusted for splits and bonuses. Refreshes: Monthly.
Test: How it was tested, and the numbers: the paid tier.
Verdict so far: Not shipped as a fund measure. 51.3% of calls were right; the same funds’ untouched holdings beat the same benchmark 51.4% of the time, so the score measures the portfolio, not the decisions. It does not persist between windows. The size-weighted version is positive for the industry as a whole (about +0.6 percentage points a year) but cannot tell one fund from another.
Started: 24 Sep 2026
Why it might matter: It is the only postulate about what a manager changed rather than what the fund owns, and the active-weight correction stops it quietly scoring momentum.
Why it might not: Monthly portfolios hide trades made and undone inside a month, and a portfolio is public only about ten days after the month it describes.
On the site: Does any of it predict?

SIF measures

Specialised Investment Funds (SIFs) are a newer, higher-minimum product regulated alongside mutual funds. They are measured, never rated, and kept out of the paper-money draw.

SIF SIFs, against what they could replaceA Specialised Investment Fund should be judged against the mutual fund it could replace, over its own dates, and by what it costs and when money can be taken out. Running

Measured: Return against the right substitute (arbitrage, balanced advantage, flexi-cap or mid- and small-cap funds) over the SIF’s own dates, net exposure from its mandate, cost against the substitute, concentration, the risk band it declares, and growth in assets. Nothing is annualised under six months.
Data: AMFI’s SIF documents and monthly NAVs; portfolios every other month. Refreshes: NAVs monthly; portfolios every other month.
Test: How it was tested, and the numbers: the paid tier.
Verdict so far: Measured and published, never rated. No verdict is possible until the funds have lived through more than one kind of market.
Started: 29 Sep 2026
Why it might matter: A new product sold at a ₹10 lakh minimum is compared with the cheaper fund it could replace, rather than with nothing.
Why it might not: Every SIF is younger than a year, and a young record flatters or punishes by the date it started.
On the site: SIFs · What the SIF label does not say

Experiment hypotheses

Questions the Fearless trade experiments ask of participants’ paper-money decisions. Answers are published as monthly aggregates only; any group smaller than five people is withheld.

The monthly reveal publishes the aggregate results. Paper money only: prices are a day late, nothing is bought or sold for real, no reward is tied to returns, and it does not train anyone for real trading.

E1 Self-picking against a blind drawChoosing a stock or fund oneself does no better than one drawn at random from the same list, over the same dates. Running

Measured: Each participant places paper money either by a blind draw from the monthly list or by choosing. Each allocation is valued at the next day’s settled price and set beside the same rupees, on the same days, in the headline index.
Data: Participants’ paper-money allocations; exchange settled prices and published NAVs, a day late. Refreshes: Daily after the close; published as a monthly aggregate.
Test: How it was tested, and the numbers: the paid tier.
Verdict so far: Running. The backtest behind the draw (163 monthly cycles, 2013–2026) found the list’s own score did not separate winners from losers, so a draw from the list is the fair yardstick. Participant results are published monthly once groups are large enough.
Started: 22 Sep 2026
Why it might matter: It is the direct test of stock-picking skill in ordinary hands, with timing held constant.
Why it might not: Paper money is not real money; people take risks with it they would not take with savings, and months of data are needed before any gap means anything.
On the site: Monthly reveal · Fearless trade experiments

E2 Reading before choosingReading more before choosing — screening, opening a fund’s measures, comparing — changes the result of the choice. Running

Measured: Research depth stamped on each self-pick at the moment it is placed, on a ladder from delegated (a blind draw) through browsed, screened, inspected and analysed to compared. Self-picks are grouped by depth and their returns set side by side.
Data: Pages and panels opened before each allocation (the experiment’s own event log); settled prices a day late. Refreshes: Daily; published as a monthly aggregate.
Test: How it was tested, and the numbers: the paid tier.
Verdict so far: Running. Too few allocations so far for any comparison.
Started: 22 Sep 2026
Why it might matter: If reading changes nothing about returns, what reading is for is understanding cost and risk, not finding winners.
Why it might not: People who read more may differ in other ways, and depth is recorded, not assigned, so this is an observation, not a controlled trial.
On the site: Monthly reveal

E3 Exits after a fallSelling after a fall costs money: what was sold goes on to do better than what the money went into. Running

Measured: Every sale with whether it was at a loss and how long it was held; gains taken against losses taken (the disposition effect: selling winners quickly while holding losers). The sold position keeps being priced after the sale.
Data: Participants’ paper-money sales; settled prices a day late. Refreshes: Daily; published as a monthly aggregate.
Test: How it was tested, and the numbers: the paid tier.
Verdict so far: Running. Too few sales so far for any comparison.
Started: 22 Sep 2026
Why it might matter: Panic selling in a fall is among the most expensive habits documented in investor behaviour, and paper money shows it without the loss.
Why it might not: A fall in paper money does not hurt the way a real one does, so behaviour here may understate the habit.
On the site: Monthly reveal

E4 Behaviour: deployment, churn, idle moneyHow money is placed — how much of the monthly amount is put to work, how often positions change, how long paper money sits unused — matters more to the result than which name is chosen. Running

Measured: Share of each month’s paper money placed, number of orders and sales, median days a position was held before sale, and paper money left unplaced.
Data: Participants’ paper-money orders and balances. Refreshes: Daily; published as a monthly aggregate.
Test: How it was tested, and the numbers: the paid tier.
Verdict so far: Running. Measured for each participant today; the aggregate appears once groups are large enough.
Started: 22 Sep 2026
Why it might matter: Cash left idle and positions changed often are the costs a statement shows most clearly and a fund page never does.
Why it might not: A single month is short, and paper money has no tax or transaction cost, so churn here is cheaper than it would be.
On the site: Monthly reveal

Read this first

Every verdict describes the past, not what comes next. Nothing here is a suggestion to buy, sell, switch or hold. Anveshan and Lineage Money are not advisers or research analysts registered with SEBI (the market regulator).