מסמך טכני מיומן המחקר · באנגלית, כפי שנכתב בשלב העבודה
Step 289: Israeli quarterly investment revisions and historical extension
The CBS 1995–2020 release provides 2012Q1–2020Q3 observations transcribed from tables 35 and 38. The archive is matched to the OECD unemployment rate via FRED, and compared with the v288 CBS 1995–2023 release. The nominal fixed investment figures in tables 35 and 38 agree for all 35 archived rows.
There are 27 overlapping quarters. The latest vintage raises the median nonhousing investment share by 2.02 percentage points of GDP, with revisions from 1.70 to 2.29 points. The median total share revision is 2.16 points. The median real GDP revision is 1.21 log points on the 100-log scale.
The extension adds eight quarters, 2012Q1–2013Q4, before the v288 sample. The raw splice uses the old values directly. The linked splice adds the median overlap revision separately to each early investment share and to log real GDP. The linked series is a sensitivity construction and its approximate intervals are conditional on the estimated link. Both specifications keep the latest vintage for all quarters from 2014Q1. The outcome excludes residential construction and still includes government nonhousing investment.
Unemployment slope for nonhousing fixed investment
Each estimate is in percentage points of nominal GDP per unemployment percentage point, as in v288. The two change horizons use the same controls and Newey–West settings as v288.
| Sample | One quarter, slope [approximate 95% interval] | Four quarters, slope [approximate 95% interval] |
|---|---|---|
| v288 recent vintage | 0.22 [−0.48, 0.92], n=38 | −0.29 [−0.89, 0.30], n=32 |
| Common dates, old vintage | 0.93 [−1.04, 2.89], n=25 | −2.55 [−3.20, −1.90], n=19 |
| Common dates, recent vintage | 1.20 [−0.72, 3.11], n=25 | −2.40 [−3.41, −1.38], n=19 |
| Extended, raw splice | 0.21 [−0.39, 0.80], n=46 | −0.33 [−0.85, 0.19], n=40 |
| Extended, median-linked splice | 0.21 [−0.42, 0.84], n=46 | −0.05 [−0.67, 0.58], n=40 |
On the same overlapping dates the two releases give similarly negative four-quarter slopes. On the longer sample the four-quarter slope is sensitive to the splice, and the extended intervals cover zero. The overlap estimates use just 19 overlapping annual changes and should not be read as structural effects. No Israeli calibration is selected. A longer single-vintage download would remove the need for this level adjustment.
Run python scripts/israel_investment_vintage_v289.py to reproduce the overlap statistics and both investment outcomes.