The reported spillovers are large, but a trader cannot observe the flow shock behind them. Ben Zeev, Ben-Ze'ev and Nathan estimate that idiosyncratic buying of Israeli central-bank bills by a handful of foreign financial institutions lifts the bill convenience yield 8.7 bps at day 480. The 2-year corporate spread falls 31.6 bps at day 380. Their ratio is about 3.6, a figure we computed from responses at different maturities and horizons; the paper does not report it. Both moves take more than a year to arrive. The identifying flows sit inside the Bank of Israel.
How does an FFI's balance sheet move the bill curve?
The money starts with a covered interest parity trade. A foreign financial institution (FFI) borrows dollars, swaps them into shekels and holds a short safe asset. The 1-month USD/ILS cross-currency basis averaged -33.70 bps over the sample, the arbitrage margin available to the trade. The asset studied is MAKAM, Bank of Israel bills with 1 to 12 month maturities. FFIs buy short government bonds (SHAHAR) as well, though MAKAM has 3.3 times the daily volume of comparable short government bonds and 85% lower price impact.
The authors' capital inflow shock is an idiosyncratic change in one large FFI's balance sheet. An earnings release or buyback, for example, gives it capacity to buy more MAKAM. In their model, the FFI puts that capacity to work with a half-life of 115 trading days. Elastic domestic holders sell as bill yields compress, and the convenience yield rises gradually. The authors argue that the wedge between policy and market rates weakens monetary transmission; mutual fund rebalancing carries the effect into corporates and equities.
They measure the convenience yield as TELBOR minus MAKAM, averaged across 1 to 12 months. Regulation links TELBOR to the overnight risk-free rate. The authors report an effectively one correlation between overnight TELBOR and the central bank's deposit rate. Foreign demand pushes MAKAM yields below that risk-free curve, widening the spread.
The underlying data comprises daily transaction-level net MAKAM flows for the 18 FFIs active from January 2017 to August 2022, giving 1,394 observations. Flows through one foreign custodian are unobserved. Primary-market auction orders account for 49.8% of FFI purchases, so secondary trading alone would miss half the demand.
To isolate a private shock, the authors construct a granular instrument (GIV). They regress each FFI's daily net inflow on its own lags, the lagged cross-currency basis, U.S. and Israeli rate changes, four OFR stress indices, TA-35 returns and Israeli CDS. They then take a size-weighted average of the residuals and subtract an inverse-variance-weighted average, removing shocks shared across institutions. Concentration gives the remaining shock force: the flow HHI is 0.47. Residual pairwise correlation falls to 5.9% from 10.9% in raw flows. A narrative check links 14 large shocks across 8 FFIs to firm-specific disclosures, including earnings windows, legal settlements, buybacks and stress-test outcomes.
The authors scale the responses to a peak increase in cumulative FFI holdings of 10pp of outstanding MAKAM after 363 days, a 3.4 standard deviation shock. The convenience yield rises 1.2 bps on impact (t-stat 3.5), then peaks at 8.7 bps after 480 trading days (t-stat 2.2). At horizon 500, the shock accounts for 39.6% of its forecast error variance. The 1-year government convenience yield rises 8.1 bps; the 10-year rises 3.9 bps. After 380 days, the 2-year investment-grade corporate spread over swaps falls 31.6 bps. The TA-35 gains 5.7% around day 390.
The funds sell, then buy elsewhere
The counterparty evidence is the paper's strongest part. In secondary trading, FFI bill purchases peak at 4pp of outstanding MAKAM, or 40% of the baseline response. Mutual fund sales reach -2.2pp at their trough, and the two response functions correlate at -91%. The funds subsequently buy 0.87pp of government bonds, 0.70pp of corporates and 0.91pp of equities. Measured at each flow's own peak, those purchases equal 41%, 37% and 43% of the funds' MAKAM sale proceeds. FFIs themselves put 1.6pp into government bonds, mainly in the 1 to 1.5 year bucket, and negligible amounts into corporates and equities.
The authors turn these flows into multipliers. For corporates, the estimate is 150 bps per pp of holdings, or 1.5 in Gabaix-Koijen units, below the 3.51 U.S. estimate of Chaudhary et al. They describe the gap as consistent with Israel's exchange-based corporate trading. For equities, the estimate is 30 against Gabaix and Koijen's 7 to 11. Mutual funds hold 3.5% of Israeli equity.
The equity estimate of 30 warrants a second identification check. The TA-35 response builds over about 390 trading days, while most foreign accumulation occurred in 2020-2022. Slow Israeli macro trends during that stretch are a plausible confound. The rebalancing data also covers secondary trading only, as the paper states. The authors say plainly that their model covers the bill market alone and leaves the cross-asset spillovers to the empirics.
Day 480, t-stat 2.2
The projections extend to 500 against a sample of 1,394 trading days. The headline horizon thus covers roughly a third of the data. Standard errors use Newey-West with truncation lag h+1 on overlapping cumulative outcomes. Sensitivity checks change lag length, end the sample at 4/11/2022, replace inverse-variance weights with equal weights and control for custodian flows. We did not find a check that changes the inference method.
Regime matters more.
The convenience yield averaged -0.4 bps through 2020, then 25.3 bps from January 2021 onward, with FFIs accumulating throughout. The authors see the acceleration predicted by their model: elastic domestic holders leave first, making later purchases more consequential. That reading is plausible. A linear impulse response nevertheless averages a near-zero regime and a 25 bp regime; its 8.7 bps peak describes neither.
Scale also matters. A one standard deviation shock produces about 1/3.4 of each impulse-response magnitude, though the variance shares stay the same. A footnote acknowledges that the comparisons with U.S. near-money elasticities put dynamic peaks beside contemporaneous estimates. The authors answer that persistence distinguishes their setting, yet the introduction headlines 8.7 bps as roughly twice U.S. elasticities. They also argue that the 2-year rise can effectively offset a 25 bp policy tightening and call the convenience-yield channel an impediment to monetary transmission. That comparison has a milder version of the same timing problem: it sets a 7.5 bp rise in the 2-year convenience yield against Kutai's 7.3 bp response of 2-year yields to a 25 bp surprise.
A shock we cannot rebuild
We could not run this on our data. The shock needs daily institution-level MAKAM flows, including primary-market auction orders, proprietary to the Bank of Israel. We also lack MAKAM, Israeli cash bond yields, TELBOR and the swap curves needed to construct the outcome variables.
The paper makes a credible market-structure case for a concentrated bill market. As a trade, it fails twice: the price responses peak 380 to 480 trading days after the shock, and the signal cannot be observed. The motivating cross-country regression (slope 1.14, R² 0.46 across 10 currencies) suggests the pattern may travel, but ten points carry little weight. A shock constructed from public auction or custody data, with the convenience-yield response still visible inside the first 100 days, would change our view.