Consumer valuation of energy-saving features of residential air conditioners with hedonic and choice models

Abstract

The promotion of energy-efficient appliances is necessary to reduce the energetic and environmental burden of the household sector. However, many studies have reported that a typical consumer underestimates the benefits of energy-saving investment on the purchase of household electric appliances. To analyze this energy-efficiency-gap problem, many scholars have estimated implicit discount rates that consumers use for energy-consuming durables. Although both hedonic and choice models have been used in previous studies, a comparison between the two models has not yet been made. This study uses point-of-sale data about Japanese residential air conditioners and estimates implicit discount rates with both hedonic and choice models. Both models demonstrate that a typical consumer underinvests in energy efficiency. Although choice models generally estimate a lower implicit discount rate than hedonic models, the latter models estimate the values of other product characteristics more consistently than choice models.

Appendix: Control variable construction

In this study, we solve the endogeneity problem between the price and the unobserved characteristics of ACs by applying the technique proposed by Hausman (1997) and Hausman and Taylor (1981). Consumers choose one AC among the ACs in the specific cooling capacity class suitable for their room. Therefore, AC markets are differentiated by cooling capacity classes. We initially estimate the prices of ACs in class c by analyzing the data of the remaining classes ($-c)$. The estimated prices (after elimination of class- and brand-specific effects) are driven by underlying costs, which provide instrumental variables that are correlated with the prices of ACs in class c, but uncorrelated with the error term $\omega _{\textit{mt}}$ in Eq. 8.

When analyzing ACs in the cth cooling capacity class, we use data of the remaining class ($-c)$ and estimate the following hedonic function:

$$\begin{aligned} \ln P_{\textit{lt}}^{-c} =\beta _0^{-c} +\beta _{\textit{CAP}}^{-c} \textit{CAP}_{{{\varvec{l}}}}^{-c} +\beta _{\textit{EC}}^{-c} \textit{EC}_{{\varvec{l}}}^{-c} +{\varvec{B}}^{-c}{\varvec{Z}}_{\textit{lt}}^{-c} +\varepsilon _{\textit{lt}}^{-c} . \end{aligned}$$

(A1)

Here, $P_{\textit{lt}}^{-c}$ is the average price of the lth AC model at period $t, \textit{CAP}_l^{-c}$ is the cooling capacity, $\textit{EC}_{{\varvec{l}}}^{-{{\varvec{c}}}}$ is annual electricity consumption, and ${\varvec{Z}}_{\textit{lt}}$ is the vector of the characteristics of the lth AC model, which include manufacture- and year-specific dummies. We use sales value as a weight. The results are presented in Table 6.

Table 6 First-stage hedonic analysis (semi-log model, Eq. A1)

Table 7 Control function derivation (Eq. A2)

Using the estimated coefficients, we calculate

$$\begin{aligned} \ln \hat{P}_{\textit{mt}}^c =\hat{\beta }_0^{-c} +\hat{\beta }_{\textit{CAP}}^{-c} \textit{CAP}_{{\varvec{m}}}^{{\varvec{c}}} +\hat{\beta } _{\textit{EC}}^{-c} EC_{{\varvec{m}}}^{{\varvec{c}}} +\hat{{\varvec{B}}}^{-c}{\varvec{Z}}_{\textit{mt}}^c. \end{aligned}$$

Plugging this into the exponential function, we estimate the expected price of the mth AC model in class c as follows:

$$\begin{aligned} \hat{P} _{\textit{mt}}^c =exp\left( {\ln \hat{P}_{\textit{mt}}^c} \right) . \end{aligned}$$

We estimate the following hedonic function:

$$\begin{aligned} P_{\textit{mt}}^c =\alpha ^{c}+\beta _E EC_{{\varvec{m}}}^{{\varvec{c}}} +{\varvec{\varGamma }}^{c} {\varvec{X}}_{\textit{mt}}^c +\zeta _{\textit{EP}} \hat{P} _{\textit{mt}}^c +\varepsilon _{\textit{mt}}^c . \end{aligned}$$

(A2)

Here, the estimated price $\hat{P} _{\textit{mt}}^c$ is used as an instrumental variable. It is correlated with the price, but uncorrelated with the error term $\omega _{\textit{mt}}$ in Eq. 7. The results are presented in Table 7. All expected price variables become positive and statistically significant at the 1% level.

Finally, we calculate the residual of this estimation as follows:

$$\begin{aligned} \mu _{\textit{mt}} =P_{\textit{mt}}^c -E[{P_{\textit{mt}}^c}]. \end{aligned}$$

Following Kim and Petrin (2010) and Petrin and Train (2010), we use this residual as a control function in Eq. 8.