Liquidity Matters: Episode 18

Liquidity Matters Episode 18 :
When does a block print become a creation or a redemption?

Each channel contributes. Taken on its own, an OTC day of €100m or more is mirrored by a single day of flow within ±10% of its size 31% of the time, an SI day 20% and an RFQ day 21%. RFQ simply carries the most volume. Looked at from the flow side, RFQ prints alone come before 20% of the creations and redemptions above €100m; adding OTC and SI lifts that to 33%. For these exact-match counts every channel is tested separately, so a trade reported in two places is never counted twice.

The side tells you which way the order goes

Block prints do not carry a side, so turnover in each channel was split into buy and sell side by estimation and the three were added up. On some days the split is decisive, on others it is not, and large days in particular tend to come out balanced or partly unassigned. We treat a day as clear when one side exceeds the other by at least half of the day’s block value.

On clear days the side carries real information. A buy-dominated block day of €20 to 50m is followed by an inflow of at least half its size 61% of the time, and by an outflow of that size only 12% of the time. A sell-dominated day of the same size is followed by an outflow 41% of the time and by an inflow 16% of the time.

Across all clear-side block days above €20m, the estimated side matches the sign of the next two days’ flow in 75% of cases, and in 77% when one side accounts for 80% or more of the day. The same test on the two days before gives 59%, most of which reflects inflows simply being more common than outflows this year. The match rate is 57% on the block day itself, 68% on the following day and 73% two days later, in line with T+2 settlement. Buy-side calls are more reliable than sell-side calls, 80% against 69%.

Most of that directional information comes from RFQ. On their own, the OTC and SI side estimates are right 65% and 62% of the time, only a few points above chance, and clear sides are rarer there. Adding them leaves direction accuracy where RFQ alone puts it, so the gain from the wider tape is in size and coverage rather than direction.

Prints and the orders behind them

The cleanest cases look like this: a block of €50m or more, almost no primary activity in the days before, and a single creation or redemption within 10% of the block one or two days later. We found 730 such pairs across 408 ETFs over the nine months, 464 of them following an RFQ print, 169 an SI print and 97 an OTC print. September alone had 71. The featured cases are discussed below; switch to all matches to filter the full set.

The clearest September case came through an SI. ESIF saw €519m printed on 17 September and a €504m redemption the next day, then €501m printed on the 18th and another €504m redeemed on the 21st. On the OTC side, SC0J printed €292m on 18 September and saw a €305m creation on the next trading day. XSXD shows the same pattern on RFQ: €411m traded on 21 September, €417m redeemed the next day.

The two UBS ETFs on 11 March show something an aggregate view hides. On the same day €159m was sold in one and €160m bought in the other, both with US equity exposure, and the primary market mirrored it the next day with a €155m redemption and a €158m creation. On 8 April the switch ran the other way, €95m in each direction, and again both legs reached the primary market, this time two days later. Switches between closely related products look like noise in total turnover. With a side estimate they become visible, and so do the orders they produce.

Prediction quality, ETF by ETF

Aggregate hit rates can hide wide differences between instruments, so we repeated the side test for each ETF with at least ten clear-side block days of €5m or more net: 360 ETFs and 6,399 days in all. The median ETF has its direction called correctly 73% of the time. On the days before the block the same ETFs score a median 56%, and 81% of them do better after the block than before it.

The harder benchmark is the naive guess. For an ETF that sees inflows on 90% of its flow days, always calling an inflow is right 90% of the time. Measured against that, the side signal adds most where flows go both ways. Among the 122 two-way ETFs, where neither direction exceeds 60% of flow days, 79% beat the naive guess. Among the 148 mixed ETFs, just over half do. Among the 90 one-way ETFs, where one direction exceeds 75%, the signal is accurate, with a median of 76%, but mostly confirms what the history already says: only 14% beat the naive guess.

Single-instrument results come with wide error bands, since most ETFs have between 10 and 30 qualifying days. 57 of the 360 beat the naive guess at a 5% significance level, a number that will grow with a longer history. The weakest group is short-dated and high yield bond ETFs, where the estimated side and the following flow often disagree. We have not yet established why, so the side signal deserves less weight in these products for now.

Inflow waves show up, rush outs less often

The data classifies each ETF by exposure region but not by sector, so the segment view groups ETFs by asset class and region. Summing net block buying across a segment and comparing it with net primary flow two days later is noisier than the instrument view, but the shape holds in the larger segments. For European equity ETFs the five-day sums move together with a correlation of 0.58, for emerging market equity 0.39.

To test the waves directly, we cut the 14 largest segments into five-day blocks and flagged blocks with an above-typical net flow. 90% of the 169 large inflow blocks were preceded by net block buying. Large outflow blocks were much rarer, 27 in nine months, and 52% of them were preceded by net block selling. That reads weaker than it is: in a year dominated by inflows, net block selling appears in only 19% of all blocks.