# Step 291: long Israeli investment panel from a single CBS release

Downloaded the quarterly [CBS Time Series API](https://www.cbs.gov.il/en/Pages/API-Time-Series.aspx) snapshot on 2026-11-10. Each quarterly series was updated 2026-10-18 and covers 1995Q1–2026Q2 (126 consecutive quarters). Series IDs are 62902 nominal GDP, 63442 chained real GDP at 2020 prices, 63061 total fixed investment, 63035 residential buildings, and 63178 fixed investment in industries. Values are NIS millions. Current-price total less residential buildings matches the directly published industries series to within NIS 0.1 million in every quarter. Summed quarterly nominal series reconcile within NIS 0.1 million with the corresponding annual series for 1995–2025 (62916, 63145, 63119 and 63179). Unemployment is the [seasonally adjusted OECD series via FRED](https://fred.stlouisfed.org/data/LRUNTTTTILQ156S), matched by calendar quarter.

Input endpoints:
- https://apis.cbs.gov.il/series/data/list?id=62902,63442,63061,63035,63178,62916,63145,63119,63179&format=json&download=false&lang=en&pagesize=1000 (raw SHA256 48e8482935e10f6e3b1f391ebe1d3d69403404a0cc0ba790cb464ddaaba09966)
- https://fred.stlouisfed.org/graph/fredgraph.csv?id=LRUNTTTTILQ156S (raw SHA256 968fbd4e74640928316113f32b2684ec19480196056c33bf59d828455c85067d)

The [CBS labour-force survey redesign](https://www.cbs.gov.il/he/publications/doclib/2022/saka0622m/intro_e_e.pdf) introduced a measurement break in 2012Q1. The break-control regression adds indicators for whether each of the two change windows crosses that quarter. Separate pre- and post-2012 regressions also allow all coefficients to differ. The v289 estimator reports the sum of the two unemployment coefficients, controlling for contemporaneous and lagged real GDP growth. Intervals are approximate Newey–West 95% intervals with four quarterly lags unless indicated.

| Sample | One-quarter nonhousing slope [interval] | Four-quarter nonhousing slope [interval] |
|---|---:|---:|
| 1995Q1–2026Q2 | −0.09 [−0.36, 0.19], n=124 | −0.54 [−0.87, −0.22], n=118 |
| Full sample with break controls | −0.10 [−0.40, 0.21] | −0.54 [−0.87, −0.21] |
| Full sample with break controls, eight HAC lags | −0.10 [−0.36, 0.16] | −0.54 [−0.84, −0.25] |
| Through 2023Q4 | −0.02 [−0.30, 0.26] | −0.52 [−0.86, −0.17] |
| Before 2012 | −0.06 [−0.49, 0.37] | −0.50 [−0.98, −0.02] |
| From 2012 | −0.05 [−0.60, 0.50] | −0.18 [−0.84, 0.49] |
| Matched v288 dates, current release | 0.13 [−0.58, 0.85] | −0.12 [−0.80, 0.56] |

Slopes are percentage points of nominal GDP per unemployment percentage point. The full-history four-quarter result survives break indicators and a longer HAC bandwidth. The post-2012 interval is wider and includes zero, and the comparison is not a formal test of a slope difference. The industry measure includes government nonhousing investment. These estimates describe historical co-movement and do not identify investment responses to automation. The full-history coefficient can be reported as a historical benchmark, alongside the post-2012 estimate. Neither is selected as a structural parameter.

Reproduce with `python scripts/israel_investment_cbs_api_v291.py`.