This repository contains the data and code used for the empirical analysis of the paper: Earthquake risk embedded in property prices: Evidence from five Japanese cities. Detailed instructions are provided to replicate all tables and figures in this paper.

1. individual_data.csv

Data from various sources have been cleaned and compiled into the data file individual_data.csv (size: 214.7 MB). This file is compressed in the zip file: individual_data.zip (size: 11.4 MB). Each row in this file represents one property transaction record, with information on the characteristics of the transaction, characteristics of the property, earthquake probabilities associated with the area that the property is located in, macroeconomic variables observed at the transaction period, and demographic characteristics of the ward that the property is located in. There are also interaction terms and derived variables.

2. Xpsi-1.csv

Short run earthquake risk data obtained through simulation is stored in Xpsi-1.csv (size: 2KB). This file contains five time series (quarterly frequency) of the short run probabilities, each column corresponding to one of the five cities included in the scope of our analysis.

3. city_range.csv

The file city_range.csv (size: 1KB) contains the chosen spatial window (for the estimation of ETAS models) corresponding to each city. Each row represents one city and the variables used in our analysis are "latMin", "latMax", "lngMin", "lngMax", corresponding to the minimum and maximum values of the latitude and longitude of the chosen spatial window (we have chosen to use rectangular spatial windows).

4. JMA_records.csv

Earthquake records collected by JMA are stored in JMA_records.csv (size: 24.15 MB). Each row of this file contains one earthquake record, with information on the time and location of the earthquake. Detailed description of the records can be found in http://www.data.jma.go.jp/svd/eqev/data/bulletin/data/shindo/format_e.txt.

For a more detailed description of the data used in our analysis as well as the data collection process, please see the accompanying data documentation (Earthquake Risk Embedded in Property Prices: Evidence from Five Japanese Cities - Data Documentation) available from https://bit.ly/3qHcTQ3. Code used for data collection and data cleaning can be provided upon request.

1. Prerequisites

R version later than 3.6.0 is needed. Additionally the packages dplyr (version 1.0.0), PtProcess (version 3.3-13), readr, R.utils and zoo are needed.

2. etas_funcs.R

This file contains the main functions used in the estimation and simulation of ETAS models. We use the R package PtProcess for the estimation and simulation.

3. R package mvecr

We have written an R package mvecr ("multivariate error components estimation in R") for the main estimation functions to replicate the analysis in the paper. To install this package, download the source file for R package mvecr (mvecr_0.3.0.tar.gz) from https://github.com/yy112/earthquake-risk and choose "Install packages from Package Archive File (.zip; .tar.gz)" in R. A manual of this package (mvecr_manual.pdf) is also included in this repository, which is more detailed about the inputs and outputs of each function in this package. A subsample of the individual_data dataset is included in this package to be used in examples.

Detailed installation instructions as well as code used to replicate tables and figures of the paper, can be found in replication_instructions.Rmd or replication_instructions.pdf.

4. replication_instructions.Rmd

This file contains all the code used for generating replication_instructions.pdf. The structure of the code is as follows:

A. Estimation and simulation of the ETAS model

• Estimation of the ETAS model, replication of the summary statistics and estimation results in Tables 48 and 50 of the Data Documentation. Computation time needed: 5 minutes.
• Simulation of short run earthquake probabilities, using the estimated ETAS parameters and historical earthquake catalog. Computation time needed: around 24 hours for 30000 Monte Carlo runs for each city. The output of the simulation is contained in Xpsi-1.csv. Figures 1 and 2 of the paper can be replicated using the data provided in Xpsi-1.csv without having to perform the simulation first.

and

B. Estimation of the multivariate error components regression model

• Characteristics of the housing dataset. Replication of Table 1 of the paper.
• Characteristics of the JSHIS long run probabilities. Replication of Table 2 of the paper.
• Main estimation results. Replication of Table 3 and Figure 3 of the paper. Computation time needed: 1 - 1.5 hours for each model.
• Sensitivity analysis regarding probability weighting functions. Replication of Table 4 and Figure 4 of the paper. Computation time needed: 1 - 1.5 hours for each model.
• Sensitivity analysis regarding other model specifications. Replication of Tables B1 - B5 of the supplementary material. Computation time needed: 1 - 1.5 hours for each model with 2 error components, 3 - 4 hours for the 3 error components model, 40 - 60 minutes for the model where individual data is grouped by nearest station instead of by district.
• Importance ordering and decomposition of risk premia. Replication of Tables C6 - C7 of the supplementary material.

These two parts can be executed independently of each other. The functions related to the estimation and simulation of the ETAS model are contained in the file etas_funcs.R. The functions related to the estimation of the multivariate error components regression model are contained in the R package mvecr.

The following output files can be found in the folder output:

1. results_long_run_only_model.csv
2. results_objective_short_run_model.csv
3. results_base_model_psi3.74.csv
4. results_SR_TK_psi1.40.csv
5. results_LR_prelec_psi3.78_gamma0.17.csv
6. results_LR_TK_psi3.77_gamma0.32.csv
7. results_attr_psi3.75.csv
8. results_noGDP_psi2.63.csv
9. results_UC_psi3.72.csv
10. results_BS_psi3.89.csv
11. results_LandUse_psi3.76.csv
12. results_noTokyo_psi1.9.csv
13. results_noNagoya_psi4.11.csv
14. results_noOsaka_psi4.04.csv
15. results_Q123_psi4.56.csv
16. results_Q4_psi3.89.csv
17. results_Tohoku_psi3.27.csv
18. results_3error_psi3.52.csv
19. results_station_psi3.41.csv

These are the results obtained from running the code contained in replication_instructions.Rmd/replication_instructions.pdf.

Ikefuji, M., Laeven, R. J., Magnus, J. R., & Yue, Y. (2021). Earthquake risk embedded in property prices: Evidence from five Japanese cities.

Ikefuji, M., Laeven, R. J., Magnus, J. R., & Yue, Y. (2021). Earthquake risk embedded in property prices: Evidence from five Japanese cities - Supplementary material.

Ikefuji, M., Laeven, R. J., Magnus, J. R., & Yue, Y. (2021). Earthquake risk embedded in property prices: Evidence from five Japanese cities - Data documentation (available from https://bit.ly/3qHcTQ3).