Spatial Dynamic Panels with Common Factors: Credit Risk in US Banking

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/stata_spxtivdfreg/data/"
use "${BASE}v113i06.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/stata_spxtivdfreg/data/"
df = pd.read_stata(BASE + "v113i06.dta")

# load every dataset at once
files = ["v113i06"]
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 + "v113i06.dta", "v113i06.dta")
df, meta = pyreadstat.read_dta("v113i06.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/stata_spxtivdfreg/data/"
df <- read_dta(paste0(BASE, "v113i06.dta"))

Overview & sources

Companion data for a Stata tutorial that replicates the empirical application of Kripfganz & Sarafidis (2025, Journal of Statistical Software 113(6)) with the spxtivdfreg package — a defactored instrumental-variables estimator for spatial dynamic panels with unobserved common factors. The panel models the non-performing-loan (NPL) ratio of 350 US commercial banks observed quarterly over 2006:Q1–2014:Q4 (36 quarters; 12,600 observations, 12,250 in the effective estimation sample), spanning the entire Global Financial Crisis. Credit risk is modelled through four simultaneous endogeneity channels — a spatial lag of NPL (ψ), temporal persistence (ρ), an endogenous regressor (INEFF, instrumented by INTEREST), and latent common factors — using a 350×350 economic-distance spatial weight matrix built from Spearman rank correlations of bank debt ratios. This page documents the bank-quarter panel (v113i06.dta); the weight matrix W.csv is a separate file (see note below).

Data sources

Cite this data

Please cite this dataset as follows.

APA

Mendez, C. (2026). Spatial Dynamic Panels with Common Factors in Stata: Credit Risk in US Banking [Data set]. https://carlos-mendez.org/post/stata_spxtivdfreg/

Kripfganz, S., & Sarafidis, V. (2025). Estimating spatial dynamic panel data models with unobserved common factors in Stata. Journal of Statistical Software, 113(6). https://doi.org/10.18637/jss.v113.i06

BibTeX

@misc{mendez2026stataspxtivdfreg,
  author       = {Mendez, Carlos},
  title        = {Spatial Dynamic Panels with Common Factors in Stata: Credit Risk in US Banking},
  year         = {2026},
  howpublished = {\url{https://carlos-mendez.org/post/stata_spxtivdfreg/}},
  note         = {Data set}
}

@article{kripfganz2025spatial,
  author  = {Kripfganz, Sebastian and Sarafidis, Vasilis},
  title   = {Estimating Spatial Dynamic Panel Data Models with Unobserved Common Factors in {Stata}},
  journal = {Journal of Statistical Software},
  volume  = {113}, number = {6}, year = {2025},
  doi     = {10.18637/jss.v113.i06}
}

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 model is a spatial dynamic panel with interactive (factor) fixed effects, estimated by defactored IV. For bank i at quarter t:

• Structural equation: NPL_it = ψ Σ_j w_ij NPL_jt + ρ NPL_i,t-1 + x_it β + α_i + λ_i' f_t + ε_it — spatial lag (ψ), temporal lag (ρ), covariates (β), bank fixed effect (α_i), and the interactive fixed effect (λ_i' f_t, the common factors).
• Spatial lag: W·NPL, the row-standardized weighted average of neighbours' NPL; w_ij > 0 when banks i, j are economic neighbours (debt-ratio Spearman correlation above the 95th percentile).
• Defactored IV: step 1 extracts and removes the common factors via principal components (the std option standardizes first); step 2 runs IV/GMM on the defactored data, instrumenting INEFF with INTEREST and lagged exogenous regressors.
• Variance share of factors: ρ_factor = σ_f² / (σ_f² + σ_e²) — fraction of residual variance due to common factors (0.335 in the full model).
• Long-run total effect: β / [(1 − ρ)(1 − ψ)] — the short-run coefficient amplified by the temporal multiplier 1/(1−ρ) and the spatial multiplier 1/(1−ψ).

The covariates x_it are the bank financial ratios INEFF, CAR, SIZE, BUFFER, PROFIT, QUALITY, and LIQUIDITY; INTEREST serves only as an excluded instrument for the endogenous INEFF.

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

• Real replication data. This is the published JSS replication panel from Kripfganz & Sarafidis (2025); it is empirical bank data, not simulated. Use it under the terms of the original replication package and cite the source article.
• The weight matrix is separate. The spatial structure lives in W.csv (a 350×350 row-standardized economic-distance matrix), not in this panel; it has no variable columns and is documented in the post (§3.3), not as a variable table here.
• INEFF is endogenous. Operational inefficiency is treated as endogenous in the NPL equation (reverse causality plus unobserved management quality) and is instrumented by INTEREST and lagged exogenous regressors — do not interpret its OLS correlation causally.
• Effective sample is smaller than 12,600. Absorbing bank fixed effects and forming the temporal lag drops the first period, leaving 12,250 observations (350 banks × 35 periods) in estimation.
• Common factors are essential for inference. Omitting them roughly doubles the temporal-persistence estimate and causes the Hansen J-test to reject (p < 0.001); the documented variables support, but do not by themselves guarantee, valid inference without the factor structure.