מסמך טכני מיומן המחקר · באנגלית, כפי שנכתב בשלב העבודה
Step 292: compare Israeli investment slopes before and after 2012
Uses the archived v291 CBS and OECD data, nonhousing fixed investment as a share of nominal GDP, and four-quarter differences with one-year lagged changes. Each regression includes current and lagged real GDP growth. Slopes sum the current and lagged unemployment coefficients. The two regressions use only changes wholly within their own survey regime.
The post-2012 minus pre-2012 slope is +0.32 points of GDP per unemployment point. The approximate 95% interval is [−0.50, +1.14] with four HAC lags (two-sided p = 0.45) and [−0.54, +1.18] with eight (p = 0.47). There are eight excluded outcome quarters between the estimation samples, so the cross-block score covariance is zero under both truncated HAC kernels. This is a comparison at the documented survey date, not a search for a break date. Longer dependence is outside the chosen covariance estimators.
| Underlying quarter window | Regression observations | Four-quarter slope [approximate 95% interval] |
|---|---|---|
| 1995Q1–2011Q4 | 60 | −0.50 [−0.98, −0.02] |
| 1995Q1–2007Q4 | 44 | −0.77 [−1.05, −0.50] |
| 2000Q1–2011Q4 | 40 | −0.50 [−1.45, 0.44] |
| 2012Q1–2026Q2 | 50 | −0.18 [−0.84, 0.49] |
| 2012Q1–2019Q4 | 24 | −0.53 [−1.71, 0.66] |
| 2012Q1–2023Q3 | 39 | +0.06 [−0.63, 0.75] |
The pre-2012 negative association is already present when the sample ends before the global financial crisis. Post-2012 point estimates are sensitive to the endpoint. The truncations are descriptive and cannot isolate the contribution of individual crises. The regime contrast is imprecise and does not establish that the slopes changed. The full-history estimate remains a descriptive benchmark alongside the recent range. No structural automation calibration is selected.
Run python scripts/israel_investment_break_v292.py.