CSI Firms' Survival Analysis

1. Define Study Scope

Industry scope: Use NACE codes 2611, 2612, and 2620 (manufacture of electronic components, loaded electronic boards, and computer equipment).

Population: All Chinese semiconductor SMEs with ≥2 employees.

Observation window: 1980–2019.

Unit of analysis: Each firm.

Outcome of interest: Firm survival duration (registration → deregistration).

2. Retrieve Firm-Level Data

Primary Source: National Enterprise Credit Information Publicity System of China (企业信用信息公示系统).

Fields: Firm name, registration date, deregistration date, employee size, branch status, accounting type, WOCO membership, and location (city-level geocode).

Filter: Keep firms classified under NACE 2611–2620 with “inactive” status by 2019.

Derived variable:

survival = deregistration_year – registration_year

3. Construct City-Level Variables

Merge firm records with city-level institutional and infrastructural data from official statistical sources:

Variable Source Computation
City Clusters China Statistical Yearbooks (firm counts per city) SMEs per 100,000 population (log-transformed)
Science Park Ministry of Science and Technology (MOST) list of National Hi-Tech Parks 1 = presence in city; 0 = none
Lead University (985) Ministry of Education of China 1 = city has 985 S&T university; 0 = no
FDI Inflow National Bureau of Statistics (NBS) 20-year mean FDI per capita (USD, logged)
Law Firms’ Density National Judicial Administration database Law firms per 100,000 population (logged)
4. Create Control Variables

Municipality: 1 = Beijing, Shanghai, Tianjin, Chongqing.

Industrial Park: 1 = city has national-level industrial park.

Branch vs. Independent: 1 = subsidiary or spin-off.

WOCO Compliance: 1 = WOCO member firm.

LF Consolidation: 1 = local-financial reporting level.

Unconsolidated Accounting: 1 = unconsolidated statements.

SIC 2611: 1 = core semiconductor manufacturing.

Coastal City: 1 = coastal province capital.

High-Speed Rail: 1 = connected city (based on China Railway Corp data).

Post-BRI Exit: 1 = exit after 2013.

Post-Tech-War Exit: 1 = exit after 2016.

5. Merge and Clean the Dataset

Merge firm-level and city-level data by city name or administrative code.

Drop duplicate firms and outliers with survival > 60 years.

Apply log transformation to continuous predictors.

Validate variable distributions and correlations (as in Table 2).

6. Verify Model Readiness

Conduct VIF diagnostic to ensure no multicollinearity (mean ≈ 1.9 acceptable).

Check for proportional hazards violation using Cox PH model.

Switch to AFT (Weibull) when proportionality fails.

7. Replicate Statistical Analysis