What investors took out of funds, and what hedge funds borrowed

Methodology

The three months are not January, February and March

The raw filing labels its flow fields MON1, MON2 and MON3. Those are the first, second and third months of the FUND'S FISCAL QUARTER, and funds keep different fiscal calendars. Form N-PORT requires a report for each month of a fiscal quarter, filed within 60 days of that quarter's end, and makes public the report for the third month; Item B.6 then asks for flows during each of the preceding three months. So MON3 is the calendar month of the report date, MON2 the month before it and MON1 the month before that. Reading MON1 as January would misdate March 2020 by two months for every fund whose fiscal quarter ends in May.

That mapping is checked rather than assumed. March 2020 reaches this panel through three groups of funds that share no members: those whose fiscal quarter ended that month, for whom it is the third month, and those whose quarters ended one and two months later, for whom it is the second and the first. If the offset were wrong for any of them, that group would be reporting an ordinary month while the others reported a run. They agree to 0.20 percentage points.

Source: SEC, Form N-PORT Data Sets | SEC, Form N-PORT (OMB 3235-0730) The three rows are disjoint sets of funds on disjoint fiscal calendars, so nothing in one row is arithmetically forced to match another. The build script refuses to write its output if they disagree by more than one percentage point. Methodology

The denominator

A rate needs net assets, and the filing reports them once, as of its own period end. For the third month of the fiscal quarter that is the true month end; for the first and second it is one or two months later, and nothing in the filing lets the earlier figures be reconstructed, because distributions paid in cash and expenses are not reported. Every rate here therefore divides a month's flow by the net assets of the filing it came from. 33.4% of the panel's rows carry a true month-end denominator. Recomputing March 2020 on those rows alone, which is a different and much smaller set of funds (6,148 funds), gives 7.16% against 7.25% for the whole month, so the choice of denominator is not what produces the episode.

The sign convention

Item B.6 asks for the total net asset value of shares sold, reinvested and redeemed, three magnitudes; the form reserves negative numbers for the gain and loss items. A minority of filers nonetheless report redemptions as a negative cash flow, and summing the two conventions together would net them against each other. The build takes the magnitude of all three components and keeps the original sign as a column so the normalisation can be audited. The two camps sit at different levels but move together. Pooled over the complete months before March 2020, the camp that files positive redeemed 2.13% of net assets and in March 2020 itself 6.13%, a multiple of 2.9; the camp that files negative went from 3.81% to 13.32%, a multiple of 3.5. The negative-filing camp is 19.4% of the panel's rows.

Coverage, and what is left out

Because each filing covers three months and a fund files once per fiscal quarter, a fund contributes at most one observation to any calendar month, and the build refuses to write the panel if one contributes two. The fund count per month is therefore a clean coverage signal, and the edges of the sample fail it. 5 months are dropped for that reason (Jul 2019, 5,606 funds; Aug 2019, 8,545 funds; Mar 2026, 10,337 funds; Apr 2026, 3,996 funds; May 2026, 3 funds), against 12,890 in February 2026. Counting fiscal-calendar cohorts alone would not have caught them all: 1 of the 5 dropped months carries all three cohorts and is trimmed on the fund count instead.

Two other exclusions are worth stating. Filings that carry no usable SEC series identifier are dropped, because the panel has no stable key for them. Fund-months missing any one of the three Item B.6 components are dropped as well, so that every row is complete and every aggregate covers the same rows. Where a fund changed its fiscal year-end and two filings overlap on a calendar month, the observation whose period end is nearer that month is kept. The build script prints the count for each of these, the net-asset share the series-identifier exclusion carries, and how many of the excluded filings report zero redemptions in all three months.

Definitions

Net flow is Item B.6.a plus Item B.6.b minus Item B.6.c: shares sold, plus shares issued for reinvested dividends and distributions, minus shares redeemed or repurchased. Reinvested distributions are counted as subscriptions because that is what they are, a distribution buying new shares, and they are reported separately so a reader can net them out. The redemption rate is Item B.6.c over Item B.1.c. Total liabilities are Item B.1.b and borrowings are Item B.2.c, amounts payable to banks and other financial institutions, within and after one year, both taken over Item B.1.c. All amounts are in US dollars, as Item B.1 requires. Nothing in the N-PORT collection is smoothed, screened by fund size, or adjusted for a fund entering or leaving the sample.

OFR Form PF aggregates

Qualifying Hedge Funds have at least $500 million in net assets and are advised by a Large Hedge Fund Adviser with at least $1.5 billion in hedge-fund assets. OFR groups the reporting funds by calendar quarter and publishes aggregate responses after rounding, screening and disclosure review. Gross assets come from Form PF question 8 and measure the market value of balance-sheet assets. Net assets come from question 9 and subtract liabilities from gross assets.

The borrowing series come from Form PF question 43 and keep OFR’s three secured-borrowing categories separate: repurchase agreements, prime brokerage and other secured borrowing. A blank value is masking, not zero, so it remains null and breaks the corresponding chart line. OFR’s historical calendar-quarter aggregates can change when amended filings arrive; this page uses the current published vintage and does not interpolate masked observations.