Mock Data assembled. Product-level sales, characteristics and prices have been collected for 50 models in China and 50 models in France (2022–2024). BLP instrument variables and the IFM index for the Chinese market (2018–2022) are computed. Estimation is in progress.
The Extended BLP Model
The commodity fetishism index IFMjt for product j in market t enters the consumer's indirect utility in the Berry, Levinsohn & Pakes (1995) framework:
Uijt = xjt βi − αi pjt + γ · IFMjt + ξjt + εijt
xjt = observable characteristics |
pjt = price |
IFMjt = spectacle index |
ξjt = unobserved quality |
εijt = i.i.d. Type I EV |
γ = spectacle premium coefficient
China PV Market 2024
23.26M
units · 40.9% NEV share
France PV Market 2024
1.72M
units · 16.9% BEV share
China Top 50 Coverage
72.1%
of total market sales
France Top 50 Coverage
66.4%
of total market sales
🇨🇳 China — Top 15 by 2024 Sales
🇫🇷 France — Top 15 by 2024 Sales
🇨🇳 China — Segment Distribution
🇫🇷 France — Segment Distribution
Product-Level Data
50 models × 2 markets
Sources: China — CPCA, Focus2Move, BestSellingCarsBlog, CarNewsChina, manufacturer websites.
France — CCFA, best-selling-cars.com, BestSellingCarsBlog, manufacturer websites.
Product j is defined at the model-year level. Prices are base MSRP in local currency (thousands).
BLP Computed Variables (2024)
100 observations
Market shares sj = salesj / Mt, where Mt is total PV registrations.
Outside good share s0 = 1 − Σsj.
Dependent variable: δjt = ln(sj) − ln(s0).
Characteristics are normalised: price/income, HP/weight, size = L×W/106.
Price/Income vs Market Share
ln(sj/s0) Distribution
Commodity Fetishism Index — Chinese Auto Market
20 models · 2018–2022
The IFM combines two components: φjt (advertising narrative ratio — share of ad time devoted to narrative elements vs. objective specs) and ψjt (hedonic residual — price premium unexplained by characteristics). A BLP-based variant uses λjt (price-to-utility ratio) instead of the hedonic residual.
φ = narrative ad ratio |
ψ̃ = normalised hedonic residual |
λ̃ = normalised BLP price/utility ratio
Foreign Luxury
0.601
Strong fetishisation
Domestic Premium
0.543
Aspirational EV strategy
JV Mass
0.407
Balanced narrative/function
Domestic Mass
0.339
Price/specs competition
IFM by Category over Time (Average)
IFM Components: φ (Advertising) vs ψ̃ (Hedonic)
IFM Rankings 2022 — Top & Bottom 5
IFM Detail Table
Low IFMHigh IFM
Brand
Model
Category
Year
φ (Ad)
ψ̃ (Hed)
λ̃ (BLP)
IFM (hed)
IFM2 (BLP)
IFM avg
Rank
Instrument Variables
Price is endogenous in the BLP model (correlated with unobserved quality ξ). Three sets of instruments are used to achieve identification:
BLP Instruments (demand-side)
IV1Own-firm avg price
IV2Own-firm avg HP
IV3Own-firm avg size
IV4Rival firms avg price
IV5Rival firms avg HP
IV6Rival firms avg size
IV7# own-firm products
IV8# rival products
Cost-Shifting Instruments
Steel HRC620 USD/t (2024)
Aluminum2,380 USD/t
Lithium98k CNY/t
Copper9,140 USD/t
CNY/USD7.10
EUR/USD0.92
CN Wage Idx124.0
FR Wage Idx118.5
Key Input Costs (2022–2024)
Full Cost Instruments Table
Variable
Unit
2022
2023
2024
Source
Steel HRC (Global)
USD/ton
950
680
620
World Steel / LME
Aluminum (LME)
USD/ton
2,705
2,255
2,380
LME
Lithium Carbonate (CN)
CNY/ton
480,000
110,000
98,000
Asian Metal / SMM
Copper (LME)
USD/ton
8,788
8,477
9,140
LME
Cobalt
USD/lb
32.50
15.20
12.80
LME / Benchmark
Nickel (LME)
USD/ton
25,600
16,400
16,200
LME
Rubber (Shanghai)
CNY/ton
12,500
12,800
14,500
Shanghai Futures
MACRO INDICATORS
CNY/USD (avg)
rate
6.73
7.08
7.10
PBoC
EUR/USD (avg)
rate
0.95
0.92
0.92
ECB
CN Mfg Wage Index
2020=100
112.3
118.5
124.0
NBS China
FR Mfg Wage Index
2020=100
108.6
114.2
118.5
INSEE
CN PV Import Tariff
%
15.0
15.0
15.0
China Customs
EU PV Tariff (std)
%
10.0
10.0
10.0
EU Commission
EU BEV Anti-Subsidy (CN)
%
0.0
0.0
17.4
EU Commission
CN NEV Tax Exemption
CNY max
30,000
30,000
30,000
MOF China
FR Bonus Écologique
EUR max
6,000
5,000
4,000
Min. Trans. Éco.
BLP Estimation
Simplified logit · 2024
Preliminary results from a simplified logit specification (no random coefficients). The full random-coefficients BLP with contraction mapping is planned using PyBLP (Conlon & Gortmaker, 2020).