AI Agents & Econometric Coding

Instrumental variables

Panel local-projection IV

Panel local-projection IV on a common calendar sample.

The task

Task statement

Read input.csv from the current directory. Its columns are unit (unit identifier), period (calendar period), y (outcome), x (endogenous regressor), and z (excluded instrument).

Estimate y responses at h=0,1,2 using 2SLS local projections. Each origin (unit,t) must have that same unit observed at calendar t-1,t,t+1,t+2. Use this common sample for every horizon. For each h, regress y_(t+h) on endogenous x_t, y_(t-1), unit fixed effects, and origin-period fixed effects. Include an intercept and omit one category from each fixed-effect set. The only excluded instrument is z_t. All included exogenous regressors instrument themselves. Report x coefficients as lp_h0,lp_h1,lp_h2. Do not treat adjacent stored rows as adjacent calendar periods. The sample count is the number of retained origin rows.

Data

Deterministic simulated data: unit, period, y, x, z

Files
input.csv

Required output

Scored estimates: lp_h0, lp_h1, lp_h2.

Reference output

TermReference estimateSample size
lp_h00.89754606481
lp_h10.16735897481
lp_h20.06590423481

A run succeeds when every scored estimate lies within 1 percent or 0.005 of its reference value, whichever is larger. Sample sizes are not scored.

Task success

Agency conditionClaude Sonnet 4.6GPT-5.6 Luna
Chatbot88.3%70.0%
Self-repair96.7%91.7%
Constrained agent100.0%93.3%
All runs
90.0% of 360 runs
By prompt condition
Zero-shot 86.7%, One-shot 93.3%
By software
Python 91.7%, R 85.8%, Stata 92.5%

Sources

Sources

About this benchmark task

Method group
Instrumental variables
Software
Stata, R, and Python
Runs
360 runs, ten in each combination of model, software, prompt, and agency condition

Open every recorded run for this task.