ueai-model-explain

Model & tools (stage 02)

Tools you may explain

Tool	Typical artefact	Where errors live
GEMPACK (tablo.exe, gemsim.exe, runGTAP.exe)	.tab (model), .cmf (command), .sl4 (solution), .har (header array)	.log and console
GTAP database 11	base data .har, parameters .prm	runGTAP window
Stata	.do, .dta, .smcl log	_rc return code
R / RStudio	.R, .Rmd, .RData	console traceback
Python (pandas, statsmodels, linearmodels)	.py, .ipynb	traceback
LaTeX (TeX Live, tabular, booktabs, estout)	.tex → .pdf	.log
Excel + Power Query	.xlsx	(light)

What you produce

A filled 02_model/model_notes.md with:

1. 工具与版本 — name and version (e.g. GEMPACK 12.0, Stata 18 SE).
2. 输入文件 — list every input artefact with relative path inside the project.
3. 核心命令 — exact command lines or .do/.R snippets the student will run.
4. 预期输出 — what files appear, what columns/coefficients to expect.
5. 常见报错 — at least 3 likely errors with diagnosis.
6. 模型假设 — for econometric models, the identifying assumption stated in plain Chinese.

Concrete patterns

GTAP shock setup (most common stage-02 case)

! tariff-shock.cmf — increase US tariff on imports of HS chapter 85 from China by 25 pp
File (new) tariffshock.sl4 ;
File (new) tariffshock.har ;
File rwsbase.har ;
File rwspara.har ;
Verbal Description = "US 25pp tariff on imports of electronics from China, 2017 base" ;
Auxiliary files = tariffshock ;
Method = Gragg ;
Steps = 3 6 9 ;
subintervals = 3 ;
automatic accuracy = yes ;
Shock tms("Electronics", "China", "USA") = 25 ;

Common errors:

• *** Could not open file XXX.har → check cd was set to the model folder.
• *** Model is not square → forgot to close a Variable or Equation block in the .tab.
• *** Newton convergence failed → reduce shock size or increase subintervals.

Stata regression (DID)

* did-tariff.do
use evidence/data/firm_panel.dta, clear
xtset firm_id year
gen post = year >= 2018
gen treat = exposed_to_301 == 1
reghdfe lnvalue treat##post, absorb(firm_id year) cluster(firm_id)
esttab using 03_results/tab1_did.tex, replace booktabs se star(* 0.10 ** 0.05 *** 0.01)

IO / completion coefficient calc

Use Leontief inverse $(I-A)^{-1}$. In Stata: mata: L = luinv(I-A). In Python:

import numpy as np
L = np.linalg.inv(np.eye(n) - A)

Output template

# 模型与工具

项目题目：{{TOPIC}}
对应课程：{{COURSE}}
研究问题：{{QUESTION}}

## 本阶段任务
记录工具命令、模型设定、报错处理和输出解释。

## AI辅助记录
- 使用时间：...
- 使用工具：...
- 使用提示词：prompts/model-gtap-gempack.md（或 model-regression.md）

### 工具与版本
- GEMPACK 12.0 / GTAP database 11
- TeX Live 2023, latexmk

### 输入文件
- 02_model/tariff-shock.cmf
- 02_model/runGTAP.bat
- evidence/data/comtrade_ch_us_2018-2024.csv

### 核心命令
```cmf
! tariff-shock.cmf
File (new) tariffshock.sl4 ;
...
Shock tms("Electronics", "China", "USA") = 25 ;

预期输出

• tariffshock.sl4 — solution
• tariffshock.har — updated headers; relevant headers: qgdp, qxw, EV
• 关注 China 的 EV、qgdp、qxw（出口数量），按部门看 Electronics、Steel

常见报错

1. Could not open file rwsbase.har → 检查工作目录；cd D:\GTAP\base 后再运行。
2. Newton convergence failed → Steps = 3 6 9 12 增加细分步数。
3. ...

模型假设

• 完全竞争 + Armington 弹性
• 全局 Walras 平衡
• 关税收入回到代表性家庭

人工核验与修改

• 数据来源或课程材料核验：
• 教师要求对照：
• 人工修改说明：

本阶段产出

• 提交文件：02_model/model_notes.md
• 运行截图/日志位置：evidence/model/

## What you DO NOT do

- Do not invent coefficients or numbers — those come from stage 03 after the model actually runs.
- Do not write a multi-equation derivation unless explicitly asked.
- If the student lacks GEMPACK access, suggest the GTAP-in-Excel or a partial-equilibrium fallback (SMART, WITS).