# Step 284: how Israeli investment responds to the Bank of Israel's rate

**Why.** Step 283 took the interest sensitivity of investment from the literature: s = 0.6 at the centre,
meaning a 1-point cut raises investment by about 6%. In the investment slump estimated from OECD data, it
is what decides the outcome: preservation ranges from 15% to 58% across the literature's range. This step
estimates it from Israeli data.

Estimates by `scripts/israel_rates_v284.py`. The data are downloaded in GitHub Actions
(`.github/workflows/israel-rates-v284.yml`) and archived in `data/processed/israel_rates_panel_v284.csv`,
with the manifest in `israel_rates_raw_manifest_v284.json`. Illustrative; not a forecast.

## Data and method

- **Data.** FRED mirrors of the OECD Main Economic Indicators, quarterly from 1995 to 2023Q3:
  - the Bank of Israel's immediate rate;
  - gross fixed capital formation (at current prices, deflated by the CPI, the only Israeli price index in
    the set);
  - real GDP (index);
  - the CPI;
  - the US federal funds rate, as an external control.
- **Shock.** The residual of a rule for the change in the policy rate. The regressors are four lags each of:
  - the rate change;
  - annual inflation;
  - GDP growth;
  - investment growth.
  The current and lagged change in the US rate are also included.
- **Response.** Local projections of cumulative investment growth over 0–12 quarters on the shock and the
  same controls. Newey–West standard errors.
- **Mapping.** The response at 8 quarters, e% per point, gives s = −e/10.

## Estimates

| Sample | Investment after 8 quarters, % per 1-point rise (s.e.) | Implied s (95% CI) |
|---|---|---|
| **1997–2023 (central)** | **+2.1 (1.9)** | **−0.21 (−0.58 to 0.17)** |
| From 2003 (inflation targeting) | +7.5 (5.0) | −0.75 (−1.73 to 0.23) |
| Two lags | +2.5 (2.0) | −0.25 (−0.65 to 0.15) |
| Up to 2019 | +1.3 (2.2) | −0.13 (−0.57 to 0.31) |

**What it shows.**

- **The estimate has the wrong sign.** It says investment rises after a rate increase, though the effect
  is not significant. This is the usual failure of this kind of identification. The Bank of Israel raises
  the rate when it expects a boom, the shekel or global conditions call for it, and the controls do not
  capture all of that. So the "shock" still contains good news about demand. The bias runs towards zero
  or a positive response.
- **The data cannot pin the sensitivity down.** The 95% interval of the central estimate runs up to
  s ≈ 0.17. It excludes the literature's central value of 0.6 but includes its low end of 0.15. Because
  the bias runs towards zero, this does not show that the true value is low. It shows only that Israeli
  aggregate data, with this method, give no support for a strong channel.
- **For the model:** step 285 runs every specification at both s = 0.6 (literature) and s = 0.15 (the low
  end, consistent with the Israeli estimate).

## Limits

- **Identification.** Recursive identification in a small open economy is weak. A credible Israeli
  estimate would need high-frequency surprises around the Bank of Israel's announcements, or a
  narrative series, neither of which is publicly available as a series.
- **Short sample.** The quarterly MEI series end in 2023Q3; the post-2003 sample has 74 quarters.
- **Nominal investment.** The CPI deflator is an approximation.