Synthetic Control with Prediction Intervals

Downloads

Each dataset is available as a labeled Stata .dta and its source file.

⇩ Download all data (ZIP)stata_codebook.do

Run stata_codebook.do in Stata once to attach long-form per-variable notes to the .dta files.

Load directly in code

Every file loads straight from GitHub (raw URLs). Swap the file name to load any dataset.

Stata

* Stata 14+ : `use` reads an https URL directly
global BASE "https://raw.githubusercontent.com/cmg777/starter-academic-v501/master/content/post/python_scpi/data/"
use "${BASE}data.dta", clear
describe
notes

Python

!pip install -q pyreadstat
import pandas as pd
BASE = "https://raw.githubusercontent.com/cmg777/starter-academic-v501/master/content/post/python_scpi/data/"
df = pd.read_stata(BASE + "data.dta")

# load every dataset at once
files = ["data"]
data = {f: pd.read_stata(BASE + f + ".dta") for f in files}

# pyreadstat (richest metadata) reads LOCAL files -> download first
import pyreadstat, urllib.request
urllib.request.urlretrieve(BASE + "data.dta", "data.dta")
df, meta = pyreadstat.read_dta("data.dta")

Copy and paste this snippet in Google Colab app. https://colab.research.google.com/notebooks/empty.ipynb

R

# R : haven::read_dta auto-downloads an https URL
library(haven)
BASE <- "https://raw.githubusercontent.com/cmg777/starter-academic-v501/master/content/post/python_scpi/data/"
df <- read_dta(paste0(BASE, "data.dta"))

Overview & sources

Companion data for a hands-on Python tutorial that applies the synthetic control with prediction intervals (SCPI) framework of Cattaneo, Feng, and Titiunik (2021) to a classic question in political economy: did German reunification in 1990 reduce West Germany's GDP per capita, and how confident can we be in the estimate? The analysis treats West Germany as the treated unit and 16 OECD countries as the donor pool, builds a synthetic West Germany from 31 pre-treatment years (1960–1990) under a simplex constraint, and constructs prediction intervals that decompose uncertainty into in-sample (weight estimation) and out-of-sample (post-treatment) components with finite-sample coverage guarantees. The estimated gap grows to roughly −\$3,465 per capita by 2003, and actual GDP falls below the 99% prediction interval in 7 of 13 post-treatment years.

Data sources

Cite this data

Please cite this dataset as follows.

APA

Mendez, C. (2026). Synthetic Control with Prediction Intervals: Quantifying Uncertainty in Germany's Reunification Impact [Data set]. https://carlos-mendez.org/post/python_scpi/

Abadie, A. (2021). Using Synthetic Controls: Feasibility, Data Requirements, and Methodological Aspects. Journal of Economic Literature, 59(2), 391–425. Cattaneo, M. D., Feng, Y., & Titiunik, R. (2021). Prediction Intervals for Synthetic Control Methods. Journal of the American Statistical Association, 116(536), 1668–1683.

BibTeX

@misc{mendez2026pythonscpi,
  author       = {Mendez, Carlos},
  title        = {Synthetic Control with Prediction Intervals: Quantifying Uncertainty in Germany's Reunification Impact},
  year         = {2026},
  howpublished = {\url{https://carlos-mendez.org/post/python_scpi/}},
  note         = {Data set}
}

@article{abadie2021using,
  author  = {Abadie, Alberto},
  title   = {Using Synthetic Controls: Feasibility, Data Requirements, and Methodological Aspects},
  journal = {Journal of Economic Literature},
  volume  = {59}, number = {2}, pages = {391--425}, year = {2021}
}
@article{cattaneo2021prediction,
  author  = {Cattaneo, Matias D. and Feng, Yingjie and Titiunik, Rocio},
  title   = {Prediction Intervals for Synthetic Control Methods},
  journal = {Journal of the American Statistical Association},
  volume  = {116}, number = {536}, pages = {1668--1683}, year = {2021}
}

Variable explorer search & filter all 11 variables

Type to filter by name or label, or use the chips to filter by type. Each row shows a mini distribution. Click a header to sort.

Cross-file variable index

Which file each variable appears in (● = present).

Construction & formulas

The synthetic control builds a counterfactual West Germany as a weighted average of donor countries, then reads the treatment effect off the post-treatment gap.

• Synthetic counterfactual: Ŷ_1T(0) = x_T' · ŵ — donor GDP values x_T at time T times the estimated weight vector ŵ.
• Simplex constraint (classic SC): w_j ≥ 0 and Σ_j w_j = 1 — a convex combination of real donors, no extrapolation. (Alternatives: lasso, ridge, OLS.)
• Treatment effect (ATT for the treated unit): τ_T = Y_1T(1) − Ŷ_1T(0) — the gap between actual and synthetic GDP in each post-1990 year. A negative gap means reunification lowered West Germany's GDP relative to its counterfactual.
• Pre-treatment fit (RMSE): RMSE = √[ (1/T₀) Σ_t (Y_t − Ŷ_t)² ] over the pre-treatment period (simplex RMSE = 0.072).
• Prediction-interval decomposition: τ̂_T − τ_T = p_T'(β₀ − β̂) [in-sample] + e_T [out-of-sample] — the in-sample term reflects finite-sample weight uncertainty (Monte Carlo, HC1 variance); the out-of-sample term e_T reflects post-treatment forecasting noise (Gaussian bound). Combining the two yields intervals with finite-sample coverage guarantees.

The datasets

Switch datasets with the tabs. Each shows the full variable dictionary plus a sortable statistics table with mini distributions and data coverage.

Expand all datasets expand to search (Ctrl/⌘+F) or print across all datasets

Known limitations & caveats

• Most columns are not used by the tutorial. Only country, year, and gdp enter the synthetic-control estimation (features=None). The seven covariate columns are retained from the source dataset for the covariate-adjustment exercise and are not part of the headline results.
• Sparse covariate coverage. Several predictors are only sparsely populated: schooling has 151 of 748 values, and invest60/invest70/invest80 are time-invariant period averages reported once per country (17 non-missing each).
• Single treated unit. With one treated unit (West Germany), the method cannot assess heterogeneity in treatment effects, and results depend on the donor-pool composition — excluding or including specific countries can shift the estimated gap.
• Cointegration assumption. The estimation sets cointegrated_data=True, assuming GDP series share a common stochastic trend; if that assumption fails, the weights may be biased.