家戶內對於電信服務之消費外部性

行政院國家科學委員會專題研究計畫 成果報告

家戶內對於電信服務之消費外部性

研究成果報告(精簡版)

計 畫 類 別 ： 個別型

計 畫 編 號 ： NSC 95-2415-H-002-039-

執 行 期 間 ： 95 年 08 月 01 日至 96 年 07 月 31 日

執 行 單 位 ： 國立臺灣大學經濟學系暨研究所

計 畫 主 持 人 ： 黃景沂

計畫參與人員： 碩士級-專任助理：石雅茹

報 告 附 件 ： 出席國際會議研究心得報告及發表論文

處 理 方 式 ： 本計畫可公開查詢

中 華 民 國 96 年 10 月 25 日

行政院國家科學委員會補助專題研究計畫成果報告

家戶內對於電信服務之消費外部性

計畫類別

個別型計畫

計畫編號

95-2415-H-002-039-執行期間

年

月

日至

年

月

日

計畫主持人

台大經濟系助理教授黃景沂

計畫參與人員

台大經濟系研究助理石雅茹

成果報告類型

精簡報告

本成果報告包括以下應繳交之附件

出席國際學術會議心得報告及發表之論文各一份

執行單位

國立台灣大學經濟學系暨研究所

中華民國

年

月

日

Abstract

I propose a game-theoretical model to estimate demand for telephone service, ac-counting for intra-household interaction among household members. Although mul-tiple Nash equilibria of subscription decisions may exist in a household, the model parameters are identified from the household-level data for any given correlation co-efficient of unobserved characteristics. Because the correlation of unobserved charac-teristics is identified only through functional form assumption, I use the correlation of observed ones to determine its magnitude. I analyze the demand in Taiwan by a semiparametric maximum likelihood estimation. The distribution of the estimated intra-household effects of cellular phone service stochastically dominates that of land-line phone service. The estimated effects of landland-line phone service are negative for all households. However, my estimation is inconclusive about the sign of effects of cellular phone consumption for most households.

我提出了一個賽局模型來估計電信需求

在此模型當中

我考慮了家戶成員之間相互的

影響。 儘管在此模型之下

使用電信服務的決定可能有多重均衡存在

不過只要給定了對於

未觀察到個人特徵的相關係數

此模型的參數即可

。 然而

此一相關係數卻無法

直接透過資料來

所以我提出的估計方法

利用可觀察到個人特徵的相關係數來決

定其大小。 我利用本研究提出的估計方法來研究台灣的電信需求

發現行動電話服務的家

戶內之效果

要高於市內電話的家戶內效果。 市內電話的家戶內效果對於每個人都是負值。

相反的

行動電話服務的家戶內之效果是正值或者是負值

並無法從資料中得到結論。

Keywords: telephone demand, semiparametric estimation, multiple Nash equi-libria, consumption externality

報告內容

1

Introduction

Standard microeconomic theory analyzes consumer behavior based on individual preferences. When more than one person lives in a household, we need to take into account the intra-household allocation of resources and consumption externalities among household members. In this study, I use a game-theoretical framework to estimate demand for telephone service and analyze intra-household effect of the subscription decision. There is a rich literature on estimating household demand for telecommunication service. Generally, these studies use household-level survey data. Nonetheless, each household is treated as a single decision-maker in the estimation. Only household heads’ individual characteristics are included in the demand estimation. This approach implicitly assumes the demand to be solely determined by household heads. Other members can influence the decision only indirectly through household-level variables.1 This assumption is unlikely to be true in most households.

As Browning, Bourguignon, Chaippori, and Lechene (1994) point out, household behavior depends on intra-household interactions unless we impose some restrictive hypotheses such as transferable utilities. They propose a collective household model: Household members bargain with each other to allocate their overall resources. Individual consumption depends on the allocation. The bargaining power depends on individual characteristics. The resource allocation must achieve Pareto efficient in the bargaining process. Using data on couples with no kids, they find that the allocation of expenditure depends on the relative incomes and ages of the couples, rejecting the hypotheses of a single decision-maker in a household. See Vermeulen (2002) for further discussions on the collective household model.

I consider a model of binary subscription choice. When there is only a single person in a household, this model reduces to a standard discrete choice model. When more than one person lives in a household, the decision would in general depend on the choices of other members. For example, if the husband has a cellular phone, the wife may have a stronger desire to own a cellular phone as well. There are several potential positive spillover effect. First, the husband can be contacted by phone even when he is away from a landline phone. This increases the wife’s demand for phone service (the direct network effect). Second, the husband’s knowledge of cellular service reduces the wife’s information cost of subscription decision (the indirect network effect). Third, when the price of a cellular-to-cellular phone call is lower than a landline-to-cellular phone call2_{, the wife pays}

less for a call from a cellular phone than from a landline phone. Besides, carriers often offer family plans which lowers the subscription fee for a second cellular phone (the price effect) On the other hand, the spillover effect may be negative if a cellular phone is a public good in a household. Then, each household member wants to be a free-rider and shares the usage of other person’s cellular phone. Whether the net effect is positive or negative is an empirical issue. However, because it is more difficult to share the usage of a cellular phone, positive effects are more likely to dominate for cellular phone service than for landline service. My empirical findings confirms this conjecture.

In the current study, I restrict my attention to households with two members. Each of the two members makes a binary choice on the subscription of telephone service. My model is similar to entry models in the industrial organization literature, such as Bresnahan and Reiss (1990). However, the effects between two firms in an entry model are always negative. The entrance of one firm reduces the profit of the other firm. Their entry decisions must be strategic substitutes. For household consumption behavior, effects may be either positive or negative. When they are positive, the decisions are strategic complements. I do not restrict the sign of intra-household effects in estimation. The sign can vary across intra-households. I will investigate how the effects affected by household characteristics.

There are two primary difficulties in the estimation. The first issue is about multiple Nash equilibria, and the second one is to identify the correlation of unobserved individual characteristics. When intra-household effect is negative, the equilibrium number of subscribers in a household is unique as in Bresnahan and Reiss (1990)’s entry model. On the contrary, multiple Nash equilibria are possible when intra-household effect is positive. To deal with multiple Nash equilibria, we can impose rules to select among these equilibria. For example, in a recent paper, Bajari, Hong, and Ryan (2007) suggest a simulation-based method to estimate the selection rule. The econometric model in this study is an extension of Tamer (2003). His approach uses a semiparametric

1

Train, McFadden, and Ben-Akiva (1987) and Train, Ben-Akiva, and Atherton (1989) only consider aggregate household income. In the estimation of demand for local telephone service under optional rate plans, Miravete (2002) includes several household-level characteristics. However, his empirical analysis only accounts for household head’s individual characteristics, but not other members’ characteristics. Many previous researches (Rappoport and Taylor, 1997; Solvason, 1997; Madden and Simpson, 1997; Duffy-Deno, 2001; Rodini, Ward, and Woroch, 2003; Economides, Seim, and Viard, 2006) use similar approach in estimating telephone demand.

2

This is the case when cellular carriers offers in-network discounts.

estimation of the data to determine the selection among multiple equilibria. I show that the parameters in the demand model can be pointwise identified for any given correlation of unobserved individual characteristics.

The unobserved characteristics of individuals within a household are unlikely to be independent. Although the correlation can be formally identified without an instrumental variable, the identification is only based on the functional form specification, not from the data. I impose the idea developed in Altonji, Elder, and Taber (2005) to gauge the within-household correlation of unobserved characteristics from the correlation of observed ones. The true value of the correlation depends on the data structure. On the one extreme, if the set of observed characteristics includes all relevant ones, the within-household correlation of unobserved characteristics is zero. On the other extreme, if observed characteristics are randomly chosen from all possible factors, the correlation of these characteristics is identical to the correlation of the unobserved ones. In general, the selection of observable characteristics is between these two extreme cases. Consequently, I require the correlation of unobserved characteristics to be bounded by the correlation of observed characteristics and zero. This requirement allows me to estimate the upper and lower bounds of intra-household effects for each household. In the next section, I introduce the econometric model and discuss the difficulties regrading demand estima-tion. In Section 3, I present my empirical results on the consumption of telephone service in Taiwan. Concludes are in the final section.

2

Econometric Model

The presence of intra-household effect implies that consumption depends on the decision of other household members. In this section, I present a static discrete response model which is an extension of the probit model. I restrict my attention to households with two members and show that, conditional on a given correlation coefficient of unobserved characteristics, the model parameters are fully identified despite the existence of multiple Nash equilibria in a noncooperative game between two household members. However, the correlation of unobserved characteristics can only be identified from functional form specification, I discuss my approach to deal with this problem in the final part of this section.

2.1 Discrete Response Model

For household i, there are two individual members j ∈ {1, 2}. All characteristics of each member are observed by both members. However, only some of the characteristics are observed by the econometrician. Denote the observed characteristics of member j by a vector xij and the unobserved characteristics by a scalar εij.

Let zi denote the vector of household-level characteristics. Since some of the individual-level characteristics

are common to both members( such as household income, residence location) the vector of individual-level characteristics xij may have overlapping elements with the vector of household-level characteristics zi. To

identify the model parameters, however, at least one of the elements in the vector xij (such as member i’s age)

is not a household-level characteristic. Furthermore, both xij and zi may contain a constant term.

Let the binary variable yij ∈ {0, 1} denote the subscription decision of individual j in household i. Let

yij = 1 if and only if the individual subscribes to a telephone service I will estimate the demand for cellular

phone service and for landline phone service separately. The demand is characterized by yij= 1 ⇔ [x

′

ijβ+ εij] + yi(3−j)[z ′

iγ] > 0, (1)

where (3 − j) is the index for the other member in the household. Let Yi = yi1+ yi2 be the total number of

subscribers in the household.

The terms in the first bracket of equation (1) represents the direct effect of consumption choice. The term in the second bracket, z′

iγ, is the intra-household effect. It is reciprocal between these two members. I normalize

the effect to be zero if the other member do not subscribe. The intra-household effect is assumed to be completely captured by observed characteristics zi.3 My model reduces to the standard probit model if the intra-household

effect vanishes (γ = 0).

The unobserved characteristics (εi1, εi2) are assumed to be jointly normally distributed, independently across

households. εi1 εi2  ∼ N0 0  ,1 ρ ρ 1  . (2) 3

In theory, I can extend the model to add an unobserved part on the intra-household effect so that it is expressed as x′ iγ+ ηi. Nonetheless, the extended model is computationally intensive. Moreover, when I estimate the model under no unobserved characteristics, the variance of the intra-household effects on consumption of cellular phone service is only 0.019 of the variable of the own effects. This suggests the magnitude of the intra-household effects is much smaller than that of the own effects. Therefore, I decide to ignore the unobserved intra-household effects for the rest of the report.

-6 (yi1, yi2) = (0, 1) (yi1, yi2) = (1, 0) (yi1, yi2) = (0, 0) (yi1, yi2) = (1, 1) (yi1, yi2) = (0, 0) or (yi1, yi2) = (1, 1) εi1 εi2 −x′ i1β −x′ i1β− z ′ iγ −x′ i2β −x′ i2β− z ′ iγ q q q q

Figure 1: Simultaneous-move non-cooperative game for positive externality

The variance of εijis normalized to one. The correlation coefficient ρ in (2) is identified only through functional

form specification, not from observed data. It cannot be accurately estimated. As a result, I apply the idea developed in Altonji et al. (2005) and use the correlation of observed characteristics to provide information about the correlation of unobserved ones. See Section 2.4 below for more details about assumptions on the unobserved factors.

2.2 Nash Equilibria

Consider a simultaneous-move non-cooperative game.4 _{This is similar to the incomplete model discussed in}

Tamer (2003). Figure 1 shows the set of equilibria for positive intra-household effect (z′

iγ> 0) conditional on

observed characteristics (xi1, xi2, zi) and unobserved characteristics (εi1, εi2). There are multiple Nash equilibria

when (εi1, εi2) ∈ (−x′i1β− z ′ iγ, −x ′ i1β) × (−x ′ i2β− z ′ iγ, −x ′ i2β). Both (yi1, yi2) = (0, 0) and (yi1, yi2) = (1, 1)

are equilibria in this region. Nonetheless, the model predicts the exact probability for (yi1, yi2) = (0, 1) and

(yi1, yi2) = (1, 0). The probability of the event (0, 0) is bounded by

Pr ({εi1< −x′i1β− z ′ iγ, εi2< −x′i2β} ∪ {εi1< −x′i1β, εi2< −x′i2β− z ′ iγ}|z ′ iγ> 0) and Pr(εi1< −x ′ i1β, εi2< −x ′ i2β|z ′ iγ> 0).

On the other hand, when the effect is negative (z′

iγ < 0), there are multiple equilibria of (0, 1) and (1, 0) if

(εi1, εi2) ∈ (−x′i1β, −x ′ i1β−z ′ iγ)×(−x ′ i2β, −x ′ i2β−z ′

iγ). (See Figure 2.) The model gives the exact probabilities

of (yi1, yi2) = (0, 0) and (yi1, yi2) = (1, 1).

Regardless the sign of intra-household effect, the exact probability of observing one subscriber in a household (Yi= yi1+ yi2= 1) for given observed characteristics (xi1, xi2, zi) can be obtain from the model.

P1(xi1, xi2, zi; β, γ, ρ) ≡ Pr (Yi= 1|xi1, xi2, zi; β, γ) = Pr (εi1< −x′i1β− z ′ iγ, εi2> −x′i2β) + Pr (εi1> −x′i1β, εi2< −x′i2β− z ′ iγ) − 1{z′ iγ< 0} Pr (−x ′ i1β< εi1< −x ′ i1β− z ′ iγ, −x ′ i2β< εi2< −x ′ i2β− z ′ iγ) , (3)

where 1{·} denotes an indicator function.

However, the exact probabilities of no subscriber (Yi = 0) and two subscribers (Yi = 2) in a household are

unknown when intra-household effect is positive because we do not know how individuals choose among multiple Nash equilibria.5 _{Without loss of generality, we only need to focus on the probability Pr(Y}

i = 0|xi1, xi2, zi)

because Pr(Yi = 2|xi1, xi2, zi) can be obtained from 1 − Pr(Yi = 0|xi1, xi2, zi) − Pr(Yi = 1|xi1, xi2, zi). The

probability of no subscriber in a household is bounded in an interval. The upper bound occurs when individuals

4

The set of Nash equilibria under a cooperative game is included in the set of Nash equilibria under a non-cooperative game. Consequently, the results under a cooperative game can be viewed as imposing an equilibrium selection rule on the results under a non-cooperative game.

5

Contrary to my model, in Bresnahan and Reiss (1990)’s entry model, the effect must be negative. As a result, the value of Yi is unique in equilibrium.

-6 (yi1, yi2) = (0, 1) (yi1, yi2) = (1, 0) (yi1, yi2) = (0, 0) (yi1, yi2) = (1, 1) (yi1, yi2) = (0, 1) or (yi1, yi2) = (1, 0) εi1 εi2 −x′ i1β− z ′ iγ −x′ i1β −x′ i2β− z ′ iγ −x′ i2β q q q q

Figure 2: Simultaneous-move noncooperative game for negative externality

always fail to coordinate their decisions in the event of multiple Nash equilibria.

P0U(xi1, xi2, zi; β, γ, ρ) ≡ Pr(εi1< −x′i1β, εi2< −x′i2β). (4)

The lower bound is achieved when individuals can perfectly coordinate. P0L(xi1, xi2, zi; β, γ, ρ) ≡ Pr(εi1< −x′i1β, εi2< −x′i2β) − 1{z′ iγ> 0} Pr(−x ′ i1β− z ′ iγ< εi1< −x ′ i1β, −x ′ i2β− z ′ iγ< εi2< −x ′ i2β). (5) 2.3 Identification

Although there are multiple Nash equilibrium in my econometric model, the parameters can be pointwise identified as long as the correlation coefficient of the unobserved characteristics ρ in (2) is known. My model is similar to Tamer (2003), which is identified when we have data on the individual decisions {(yi1, yi2)}. However,

the data set that I use only reports the aggregate decision in a household (Yi = yi1+ yi2), not individual choices.

The following theorem is an extension of Theorem 1 in Tamer (2003).

Theorem 1. _{Suppose that there exists a regressor of individual characteristics (xi1k, xi2k) with xi1k, xi2k ∈ z/ i} andβk6= 0 and such that the conditional distribution of xi1k|x−i1k has an everywhere positive Lebesgue density

where x−i1k= (xi11, . . . , xi1,k−1, xi1,k+1, . . . , xi1K). Then the parameter vectors, β and γ, are identified for any

given covariance matrix of unobserved characteristics if the matrices {xi1 : i = 1 . . . , N }, {xi2: i = 1, . . . , N },

and{zi: i = 1, . . . , N } have full rank.

2.4 Unobserved Characteristics

As I mentioned at the beginning of this section, the observed data cannot truly identify the covariance matrix of the unobserved characteristics (2). In fact, the correlation coefficient ρ can not be separately identified from the mean of consumption externality. To demonstrate this, consider the case z′

iγ = γ0 is a constant for all

household.

For positive externality γ0> 0,

P1(xi1, xi2, zi; β, γ0, ρ) = Pr(x′ i1β+ γ0+ εi1< 0, x′i2β+ εi2> 0) + Pr(x′i1β+ εi1> 0, x′i2β+ γ0+ εi2< 0) = Z −x′i1β−γ0 −∞ Z x′i2β −∞ e− ε2_{i1 +ε}2_{i2+2ρεi1 εi2} 2(1−ρ2 ) 2πp1 − ρ2 dεi2dεi1+ Z x′i1β −∞ Z −x′i2β−γ0 −∞ e− ε2_{i1 +ε}2_{i2+2ρεi1 εi2} 2(1−ρ2 ) 2πp1 − ρ2 dεi2dεi1. 4

Compute the partial derivatives with respect to ρ and γ0, respectively.6 ∂P1(xi1, xi2, zi; β, γ0, ρ) ∂ρ = − e−(x′i1β+γ0) 2 +(x′ i2β)2 −2ρ(x′i1β+γ0)(x′i2β) 2(1−ρ2 ) πp1 − ρ2 < 0. ∂P1(xi1, xi2, zi; β, γ0, ρ) ∂γ0 = − Z x′i2β −∞ e−(x′i1β+γ0) 2 +ε2i2−2ρ(x′i1β+γ0)εi2 2(1−ρ2 ) 2πp1 − ρ2 dεi2− Z x′i1β −∞ e−(x′i2β+γ0) 2 +ε2i1−2ρ(x′i2β+γ0)εi1 2(1−ρ2 ) 2πp1 − ρ2 dεi1< 0.

According to the implicit function theorem, the above two inequations imply dγ0 dρ = − ∂P1(xi1,xi2,zi;β,γ0,ρ) ∂ρ ∂P1(xi1,xi2,zi;β,γ0,ρ) ∂γ0 < 0 (6)

for any given (xi1, xi2, zi).

For negative externality γ0< 0, I can similarly obtain

∂P1(xi1, xi2, zi; β, γ0, ρ) ∂ρ = − ∂ Pr(Yi= 0|xi1, xi2, zi) ∂ρ − ∂ Pr(Yi= 2|xi1, xi2, zi) ∂ρ < 0, and ∂P1(xi1, xi2, zi; β, γ0, ρ) ∂γ0 = −∂ Pr(Yi= 0|xi1, xi2, zi) ∂γ0 −∂ Pr(Yi= 2|xi1, xi2, zi) ∂γ0 < 0. As a result, the implicit function theorem also implies dγ0/dρ < 0 for any given (xi1, xi2, zi).

As Altonji et al. (2005) suggest, if the selection of observed characteristics is completely random from the set of all relevant factors, the correlation coefficient of observed characteristics is identical to that of unobserved characteristics,

ρ = Cov(x′ i1β, x

′

i2β) (7)

On the contrary, if all relevant factors are included in the set of observed characteristics, the unobserved characteristics are purely random noises. In the latter case, the correlation among two household members is zero,

ρ = 0. (8)

The reality is likely to lie between the above two extreme cases. Since I have tried to include the most important variables in the set of regressors xij, the correlation of the unobserved characteristics is likely to have

the same sign as that of observed characteristics, but the magnitude is smaller.7

0 ≤ ρ ≤ Cov(x′ i1β, x

′

i2β). (9)

Equation (6) shows that there is a negative relationship between the correlation coefficient ρ and the intra-household effect. Therefore, the intra-intra-household effect estimated under Condition (7) is an lower bound of the true value while the effect estimated under Condition (8) is an upper bound.8

2.5 Semiparametric Maximum Likelihood Estimator

If consumption effect is negative, I know the exact probability of the events {Yi = 0}, {Yi= 1}, and {Yi= 2}

conditional on the observed characteristics. Consequently, the usual likelihood can be computed. On the contrary, the exact probabilities of {Yi = 0} and {Yi = 2} are unknown when externality is positive. I use a

semiparametric maximum likelihood estimator, extended from Tamer (2003), to obtain the parameters in the demand model. Define the conditional probability of the event {Yi= 0} for observed characteristics (xi1, xi2, zi)

6

The computational details are provided in an appendix available from the author. 7

From the viewpoint of modeling, it is possible to have negative correlation and to use the condition Cov(x′

i1β, x′i2β) < ρ < 0. However, in the estimation, I find the correlation to be positive.

8

If the joint distribution of unobserved characteristics (εi1, εi2) is totally unrestricted, it is bounded by max{Φ(εi1) + Φ(εi2) − 1, 0} ≤ F (εi1, εi2) ≤ min{Φ(εi1), Φ(εi2)}, where Φ is the distribution function of standard normal (Hoeffding, 1940). Although these bounds are tight, they are too wide to obtain meaningful parameter estimates.

as

H(xi1, xi2) = Pr (Yi= 0|xi1, xi2, zi) .

When H is known, I can write down the likelihood and the parameters (β, γ) are estimated by maximizing the logarithm of the likelihood function. For a random sample with size N ,9 _{the logarithm of the likelihood}

function is L(β, γ, ρ; H) = 1 N X i  1_{[Yi= 0] log(H(xi1, xi2, zi))} +1[Yi= 1] log (P1(xi1, xi2, zi; β, γ, ρ)) +1[Yi= 2] log  1 − H(xi1, xi2, zi) − P1(xi1, xi2, zi; β, γ, ρ)  (10) The unknown function H(xi1, xi2, zi) represents the probability of observing no subscriber in a household

in the event of multiple Nash equilibria. From equations (4) and (5), we know H(xi1, xi2, zi) is bounded

by the closed interval [PL

0(xi1, xi2, zi; β, γ, ρ), P0U(xi1, xi2, zi; β, γ, ρ)], but the model cannot predict the exact

probability. To overcome this difficulty, Tamer (2003) suggest to approximate the unknown function by a kernel regression of the event {Yi = 0} on (xi1, xi2, zi). Since H(xi1, xi2, zi) is bounded by P0L(xi1, xi2, zi; β, γ, ρ) and

PU

0 (xi1, xi2, zi; β, γ, ρ), I truncate the result of the kernel regression by the upper and lower bounds and denote

the value by ˆH(xi1, xi2, zi; β, γ, ρ). Replace H in the likelihood (10) by ˆH. I can obtain a consistent estimate

of (β, γ) for any given value of the correlation ρ. The asymptotic variance of the estimate can be computed from the score and Hessian of the log likelihood (10).10

I first estimate the model under Condition (7) ρ = 0 to obtain an upper bound of intra-household effects. Then, I estimate the model subject to Condition (8) ρ = Cov(x′

i1β, x ′

i2β) to find a lower bound of the effect.

Although the model is described under a simultaneous non-cooperative game, it actually includes coopera-tive game as a special case. If all households can coordinate their consumption decisions, the kernel estimation of H(xi1, xi2, zi) will converge to P0L(xi1, xi2, zi) in probability. Similarly, if individuals make decisions

sequen-tially, then the subsame-prefect equilibrium is also included in the set of Nash equilibria under a simultaneous non-cooperative game.

3

Empirical Results

3.1 Data

I use cross-sectional data from Taiwan, the 2003 Survey of Family Income and Expenditure. This survey was conducted by the Directorate-General of Budget, Accounting and Statistics in early 2004. It adopts a stratified two-stage sampling method with counties and cities as subpopulations. The universal sampling rate is 0.20%, which is 13,681 households. Because young kids are unlikely to make their own decisions and they are unlikely to use telephones, young kids are not counted as household members in my empirical work. I define young kids as people who are less than 6 years old. The estimation results do not change much for different definition of young kids. Based on this age criterion, there are 3,489 households with two members.

I only observe the total numbers of cellular phones and landline phones in a household. When the total is zero, obviously neither member subscribes to the phone service. When it is one, only one member in the household choose to subscribe, and the other member does not. When there is more than one phone, I assume that both individuals choose to have one. In my data, 3% of two-member households own more than two cellular phones, and 1% of households have more than two landline phones.

3.2 Demand for Cellular Phone Service

I first estimate the model under the assumption that the correlation of the unobserved characteristics within a household is zero. I will obtain an upper bound estimate of the distribution of intra-household effects. Then, I estimate the lower bound of the distribution by imposing the constraint Cov(x′

i1β, x ′

i2β) = ρ. 9

The survey data I use to perform estimation is not a random sample. Therefore, I need to adjust for the sampling weights in my calculation. To ease the exposition, however, I present the estimator for a random sample in this section.

10

These two matrices are asymptotically equal for a random sample. However, for the survey data, I need to account for sampling weights in the estimation and the matrices are different.

−0.350 −0.3 −0.25 −0.2 −0.15 −0.1 −0.05 0 0.05 0.1 0.02 0.04 0.06 0.08 0.1 0.12 0.14

Marginal Effect of Intra−Household Effect

Probability

Figure 3: Histogram of the estimated intra-household effects for cellular phone service under zero correlation of unobserved characteristics

3.2.1 Zero Correlation of Unobserved Characteristics

The parameter estimates for the choice of cellular phone service subscription under zero correlation are presented in Table 1. In addition to the household-level variable reported in Table ??, I also construct six additional household-level characteristics by averaging and differencing individual-level variables: average age, average education years, average employment status, age difference, education difference, and income difference. I report four different specifications. In the first two columns, differences of individual characteristics within a household are not included. In Column (A), I also exclude the regional dummy variables (Central and South) in the set of observed characteristics. In Column (D), the full set of household-level characteristics is used in the estimation. Some of the covariates are insignificantly different from zero. These covariates are excluded in the estimation for Column (C).

According to likelihood-ratio tests, Column (A) and (B) are both significantly different from Column (D) at the 95% level, suggesting the differences of individual characteristics within a household have significant effects on demand for cellular phone service. On the contrary, no significant difference between the last two columns. I will use the estimation results in Column (C) as the benchmark case.

The magnitude of intra-household effect can be expressed as the marginal effect of one household member’s subscription decision on the other member. For member j in household i, the marginal effect is

Pr([x′

ijβ+ εij] + [z′iγ] > 0) − Pr([x ′

ijβ+ εij] > 0).

Based on the estimated parameters, ˆβand ˆγ, I compute the marginal effect for each individual. Figure 3 shows the distribution of the estimated marginal effects due to the externalities among household members. 63.56% of the estimated effects are positive. On average, the effect increases subscription by 0.93 percentage points. It has standard deviation 4.75 percentage points. When the other household member chooses to subscribe, its average effect is equivalent to the effect caused by a 46,766 TWD (equal to 1,358 USD) increase in individual annual income. Although the average effect is close to zero, the variation of the effect is large among households.

As Figure 3 shows, intra-household effects vary across households. Estimate of the vector γ differ significantly from zero at 5% level for several elements. The effects increase when the number of employed person in a household increases, meaning the network effect of cellular phone usage is stronger when more household members are employed. On the contrary, the effects decrease in the number of young kids and the age difference in a household. People tend to share the usage of a single cellular phone when they have more kids. In addition, the effects are higher for household living in a city. The network effect appear to be stronger in cities. However, there is no significant difference across the regions: North, Central, and South.

3.2.2 Positive Correlation of Unobserved Characteristics

When I assume zero correlation of unobserved factors to obtain Table 1, the correlation coefficient of observed characteristics among individuals in a household is 0.639. This suggests that assuming unobserved characteristics to be uncorrelated among household members might be too restrictive. In this section, I impose the constraint (7) : The correlation coefficient is the same for observed and for unobserved characteristics, Cov(x′

i1β, x ′ i2β) =

Table 1: Estimation results for cellular phone service under zero correlation (A) (B) (C) (D) Characteristics β γ β γ β γ β γ constant 0.393 -0.404 0.430 -0.416 0.319 0.008 0.401 -0.072 (0.135) (0.189) (0.146) (0.193) (0.126) (0.122) (0.143) (0.229) Household Income 0.242 0.007 0.228 0.016 0.237 0.261 0.054 (0.121) (0.111) (0.122) (0.112) (0.083) (0.147) (0.134) [0.061] [0.002] [0.058] [0.004] [0.061] [0.067] [0.014] Town -0.049 -0.196 -0.039 -0.193 -0.229 -0.035 -0.199 (0.054) (0.077) (0.051) (0.078) (0.076) (0.048) (0.078) [-0.013] [-0.050] [-0.010] [-0.049] [-0.058] [-0.009] [-0.051] Rural -0.078 -0.166 -0.058 -0.165 -0.224 -0.057 -0.170 (0.085) (0.132) (0.073) (0.134) (0.128) (0.071) (0.137) [-0.020] [-0.042] [-0.015] [-0.042] [-0.056] [-0.015] [-0.043] Central 0.014 -0.037 -0.011 0.017 -0.045 (0.052) (0.071) (0.043) (0.045) (0.071) [0.004] [-0.009] [-0.003] [0.004] [-0.011] South -0.097 0.034 -0.093 -0.097 0.027 (0.034) (0.060) (0.031) (0.032) (0.059) [-0.024] [0.008] [-0.024] [-0.025] [0.007] Number of Kids 0.096 -0.163 0.090 -0.161 0.115 -0.240 0.103 -0.222 (0.077) (0.080) (0.077) (0.080) (0.075) (0.077) (0.080) (0.085) [0.024] [-0.041] [0.023] [-0.040] [0.030] [-0.061] [0.026] [-0.056] Average Age 0.475 0.491 0.103 (0.268) (0.264) (0.293) [0.120] [0.123] [0.026] Avg. Education 0.193 0.182 0.160 0.169 (0.108) (0.107) (0.099) (0.106) [0.049] [0.046] [0.040] [0.043] Avg. Employment 0.213 0.220 0.162 0.147 (0.089) (0.089) (0.082) (0.100) [0.054] [0.055] [0.041] [0.037] Age Difference -0.747 -0.809 (0.234) (0.268) [-0.189] [-0.205] Edu. Difference 0.099 (0.100) [0.025] Inc. Difference -0.088 (0.133) [-0.022] Gender -0.033 -0.033 -0.053 (0.074) (0.073) (0.081) [-0.008] [-0.008] [-0.014] Age -2.853 -2.867 -2.715 -2.792 (0.146) (0.150) (0.147) (0.160) [-0.722] [-0.725] [-0.697] [-0.714] Education 0.651 0.664 0.649 0.653 (0.055) (0.062) (0.053) (0.059) [0.165] [0.168] [0.167] [0.167] Employment 0.298 0.298 0.289 0.285 (0.054) (0.054) (0.045) (0.048) [0.080] [0.089] [0.078] [0.077] Individual Income 0.749 0.746 0.774 0.696 (0.177) (0.173) (0.134) (0.217) [0.190] [0.189] [0.199] [0.178] Likelihood -2539.735 -2537.018 -2533.284 -2531.128

Notes:_{Standard errors are in parentheses. Marginal effects are in square brackets, which are computed as average derivatives} of the subscription probability except for for dummy variables, whose effects are evaluated for a move from 0 to 1. The sample size is 3489 households or 6978 individuals.

−0.70 −0.6 −0.5 −0.4 −0.3 −0.2 −0.1 0 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08

Marginal Effect of Intra−Household Effect

Probability

Figure 4: Histogram of the estimated externalities for cellular phone service under positive correlation of unobserved characteristics

Cov(εi1, εi2) = ρ.

The estimated parameters under the constraint on the correlation of unobserved characteristics are presented in Table 2. I estimate the model under the sets of covariates in Column (A) and (C) of Table 1. The distribution of the estimated marginal intra-household effect under specification (C) is illustrated in Figure 4. It has mean −13.92 percentage points and standard deviation 7.87 percentage points. All of the estimated effects are negative.

The estimation under the assumption of random selection of observed characteristics in this subsection gives an lower bound estimate of intra-household effects. When I compare the result with the upper bound estimated zero correlation of unobserved characteristics, the range between these two bounds is (−0.1392, 0.0093). There-fore, I am inconclusive about the sign of average intra-household effect for cellular service demand. Nonetheless, the coefficients of intra-household effect are qualitatively robust except the constant term γ0. The magnitude of

the marginal effects of these coefficients does not change much as well. Ignoring the correlation of unobserved characteristics results in an higher estimate of the constant term for intra-household effect γ0. The effects on

the estimation of other covariates are much smaller in the current study.

As I discussed in Subsection 2.4, the true correlation of unobserved characteristics ρ is likely to lie between the two extreme cases that I have estimated above. When 0 ≤ ρ ≤ Cov(x′

i1β, x ′

i2β), the true distribution

of consumption externalities lies between the two estimated distributions shown on Figure 3 and Figure 4, respectively. Consequently, it is inconclusive to determine the sign of consumption externality of cellular phone service for most households. Nonetheless, at least 36 44% of them are estimated to be negative.

3.3 Demand for Landline Phone Service

Next, I apply the same estimation approach to the demand for landline phone service. Column (D) in Table 3 use the same set of covariates as Column (D) in Table 1. In Column (E), I exclude covariates which has t-ratio less than 1 in the estimation of Column (D). There is no significant difference between these two specifications. Therefore, the following discussion of estimation result is based on Column since it contents fewer covariates.

The estimations are done under the assumption of zero correlation of unobserved characteristics. As a result, the estimated intra-household effects are upper bounds for landline phone service. The distribution of the estimated marginal intra-household effects is illustrated in Figure 5. Its mean is −55.23 percentage points and standard deviation is 15.25 percentage points. All of the estimated effects are negative. The estimated correlation coefficient of observed characteristics is 0.5401. If the true correlation of unobserved characteristics is also positive, the estimated distribution in Figure 5 must first-order stochastically dominate the true distribution. Consequently, I can conclude that intra-household effect of landline phone service is negative for all households in my sample. On average, an individual’s decision to subscribe to landline phone service reduces the other household member’s probability of subscription by at least 55.23 percentage points.

The intra-household effect for landline phone demand is higher for household in the North region, but it decrease in household income and average age. Interestingly, similar to the findings in Browning et al. (1994), the differences of characteristics within a household have significant role in the intra-household effect. Difference in education has a negative effect on intra-household effect, while difference in individual income has a positive effect.

Table 2: Parameter estimates under random selection of observed characteristics (A) (C) Characteristics β γ β γ constant 0.737 -1.053 0.665 -0.782 (0.025) (0.230) (0.022) (0.148) Household Income 0.666 -0.290 0.382 (0.026) (0.127) (0.038) [0.160] [-0.066] [0.094] Town -0.080 -0.229 -0.257 (0.007) (0.101) (0.087) [-0.019] [-0.052] [-0.059] Rural -0.092 -0.243 -0.259 (0.010) (0.187) (0.151) [-0.022] [-0.055] [-0.059] Central -0.017 (0.008) [-0.004] South -0.112 (0.008) [-0.028] Number of Kids 0.218 -0.306 0.225 -0.328 (0.008) (0.053) (0.009) (0.050) [0.052] [-0.070] [0.055] [-0.075] Average Age 0.228 (0.276) [0.052] Average Education 0.411 0.250 (0.142) (0.112) [0.094] [0.057] Average Employment 0.395 0.296 (0.110) (0.092) [0.090] [0.068] Age Difference -0.299 (0.263) [-0.068] Gender -0.018 (0.011) [-0.004] Age -3.077 -2.944 (0.025) (0.024) [-0.741] [-0.722] Education 0.593 0.680 (0.010) (0.008) [0.143] [0.167] Employment 0.241 0.263 (0.010) (0.010) [0.060] [0.067] Individual Income 0.408 0.588 (0.042) (0.050) [0.098] [0.144] Likelihood -2575.601 -2569.568 Correlation ρ 0.7962 0.7259

Notes: _{Standard errors are in parentheses. Marginal effects are computed} as average derivatives of the subscription probability except for for dummy variables, whose effects are evaluated for a move from 0 to 1. The sample size is 3377 households or 6754 individuals.

Table 3: Estimation results for landline phone service under zero correlation (D) (E) Characteristics β γ β γ constant -1.281 0.231 -1.177 0.098 (0.333) (0.389) (0.259) (0.264) Household Income 1.230 -0.557 1.322 -0.645 (0.285) (0.256) (0.271) (0.229) [0.223] [-0.172] [0.243] [-0.199] Town -0.096 -0.040 (0.123) (0.151) [-0.018] [-0.012] Rural 0.147 -0.074 (0.225) (0.303) [0.025] [-0.023] Central 0.232 -0.386 0.226 -0.394 (0.135) (0.152) (0.123) (0.139) [0.040] [-0.114] [0.039] [-0.116] South 0.165 -0.345 0.158 -0.349 (0.108) (0.121) (0.100) (0.113) [0.029] [-0.104] [0.028] [-0.105] Number of Kids 0.091 -0.165 -0.076 (0.105) (0.113) (0.047) [0.016] [-0.051] [-0.024] Average Age -2.680 -2.487 (0.479) (0.337) [-0.830] [-0.769] Avg. Education -0.081 (0.184) [-0.025] Avg. Employment -0.020 (0.184) [-0.006] Age Difference 0.286 0.263 (0.265) (0.258) [0.089] [0.081] Edu. Difference -0.313 -0.316 (0.105) (0.105) [-0.097] [-0.098] Inc. Difference 0.281 0.283 (0.112) (0.108) [0.087] [0.088] Gender 0.345 0.355 (0.131) (0.133) [0.062] [0.065] Age 2.234 2.034 (0.420) (0.312) [0.405] [0.373] Education 0.520 0.480 (0.159) (0.075) [0.094] [0.088] Employment 0.053 (0.152) [0.010] Individual Income -0.771 -0.775 (0.269) (0.274) [-0.140] [-0.142] Likelihood -1347.186 -1349.943

Notes: _{Standard errors are in parentheses. Marginal effects are in square} brackets, which are computed as average derivatives of the subscription prob-ability except for for dummy variables, whose effects are evaluated for a move from 0 to 1. The sample size is 3489 households or 6978 individuals.

−0.90 −0.8 −0.7 −0.6 −0.5 −0.4 −0.3 −0.2 −0.1 0 0.01 0.02 0.03 0.04 0.05 0.06

Marginal Effect of Intra−Household Effect

Probability

Figure 5: Histogram of the estimated externalities for landline phone service under zero correlation of unobserved characteristics

4

Conclusion

I empirically analyze intra-household effect on the demand for telephone services. Because of the effect, it is possible to have multiple equilibria in a non-cooperative game. Nonetheless, the model is fully identified from household-level data as long as the correlation coefficient of unobserved characteristics is given. Since the correlation cannot be directly identified from the data expect through functional form assumption, I restrict its value to be between zero and the correlation of observed characteristics, based on the idea of selection on observed and unobserved characteristics. This restriction allows me to obtain upper and lower bounds of intra-household effects. I use a semiparametric maximum likelihood estimator to recover the demand for cellular phone service in Taiwan. The sign of intra-household effect of cellular phone service is inconclusive for most households, but the effect of landline phone service is negative for all households.

In the current study, I consider demand for cellular phone service and for landline phone service separately. An interesting extension is to estimate demand for these two services jointly. Another important future work is to include households with more than two individuals. Contrary to the two-member case, the exact probability of any observed event is unknown due to multiple equilibria. The parameters are only partially identified by inequalities.

References

Altonji, J. G., T. E. Elder, and C. R. Taber (2005). Selection on observed and unobserved variables: Accessing the effectiveness of Catholic schools. Journal of Political Economy 113, 151–184.

Bajari, P., H. Hong, and S. Ryan (2007). Identification and estimation of a discrete game of complete informa-tion. Mimeo. University of Minnesota, Stanford University, and Massachusetts Institute of Technology. Bresnahan, T. F. and P. C. Reiss (1990). Entry in monopoly markets. Review of Economic Studies 57, 531–553. Browning, M., F. Bourguignon, P.-A. Chaippori, and V. Lechene (1994). Income and outcomes: A structural

model of intrahousehold allocation. Journal of Political Economics 102, 1067–1096.

Duffy-Deno, K. T. (2001). Demand for additional telephone lines: An empirical note. Information Economics and Policy 13, 283–299.

Economides, N., K. Seim, and B. V. Viard (2006). Quantifying the benefits of entry into local phone service. Mimeo. New York University and Stanford University.

Hoeffding, W. (1940). Masstabinvariate Korrelationstheorie. Schriften des Mathematischen Instituts und des Institutes F¨ur Angewandte der Mathematik der Universit¨at Berlin 5, 179–233.

Madden, G. and M. Simpson (1997). Residential broadband subscription demand: an econometric analysis of australian choice experiment. Applied Economics 29, 1073–1078.

Miravete, E. J. (2002). Estimating demand for local telephone service with asymmetric information and optional calling plans. Review of Economic Studies 69, 943–971.

Rappoport, P. N. and L. D. Taylor (1997). Toll price elasticities estimated from a sample of U.S. residential telephone bills. Information Economics and Policy 9, 51–70.

Rodini, M., M. R. Ward, and G. A. Woroch (2003). Going mobile: Substitutability between fixed and mobile access. Bell Journal of Economics and Management Science 27, 457–476.

Solvason, D. L. (1997). Cross-sectional analysis of residential telephone subscription in canada using 1994 data. Information Economics and Policy 9, 241–264.

Tamer, E. (2003). Incomplete simultaneous discrete response model with multiple equilibria. Review of Economic Studies 70, 147–165.

Train, K. E., M. Ben-Akiva, and T. Atherton (1989). Consumption patterns and self-selecting tariffs. Review of Economics and Statistics 71, 62–73.

Train, K. E., D. L. McFadden, and M. Ben-Akiva (1987). The demand for local telephone service: A fully discrete model of residential calling patterns and service choices. RAND Journal of Economics 18, 109–123. Vermeulen, F. (2002). Collective household models: Principles and main results. Journal of Economic

Sur-veys 16, 533–564.

計畫成果自評

The methods used in this study is not exactly the same as my original proposal. During the progross of this study, I realized the importance of the correlation on unobserved characteristics in my estimation. Therefore, I turned more attention to this new direction.

The paper has been accepted to several conferences, such as North American Summer Meeting of the Econometric Society and Far Eastern Meeting of the Econometric Society.

The working paper version of my study has been posted on the SSRN website http://ssrn.com/abstract= 991772. I will submit this paper for publication by the end of October.

出國報告書

會議名稱: 2007 North American Summer Meetings of the Econometric

Society

會議時間: 2007.6.21-2007.6.24

會議地點: Duke University, Durham NC, USA

學術交流訪問: 2007.6.26-2007.7.6 Northwestern University, Chicago IL,

USA

國立台灣大學經濟學系

助理教授

一、目的

Econometric Society 是國際上最重要的經濟學會之一，每年夏天都會在北美地區

舉辦一場經濟學研討會。此次參與的目的，除了發表了一篇文章 Intra-Household

Consumption Externalities for the Cellular Phone Service 之外，並且聽取了許多相

關領域的最新研究。研討會之後，轉往位在芝加哥的 Northwestern University 訪

問，該校為在產業組織(Industrial Organization)領域方面的研究，在美國居於領導

地位。

二、與會過程

在經歷了班機因故延誤與取消、重新劃位等等波折之後，在 2007 年 6 月 20 日傍

晚終於順利抵達 Duke University。在完成 check-in 手續，住到該校的宿舍之後，

才發現校園內的所有商店都在七點就打烊了，但是校外的商店卻是超出步行範

圍。所以當晚只好餓肚子了。(受限於行政院的生活日支費標準，無財力負擔校

外的大會指定飯店的住宿費用，否則至少有飯店的餐廳可以吃。)

6 月 21 日開始參加會議，會議在該校商學院 Faqua School of Business 進行。場

地的硬體設備相當完善，議程的安排也算是相當流暢。

在接下來的這四天的會議當中，我依序參加了以下的 parallel sessions，聽到了相

當多有興趣的文章：

1. Pricing, Advertising, and Entry/Exit

De

l

t

a

/

Cont

i

ne

nt

a

l

/

Nor

t

hwe

s

t

Code

s

ha

r

e

Al

l

i

a

nce”by Philip Gayle, Kansas

State University

ii.

“

Ta

r

ge

t

e

d Adve

r

t

i

s

ing: The Role of Subscriber Characteristics in Media

Ma

r

ke

t

s

”by Ambarish Chandra, University of British Columbia

iii.

“

Pr

e

da

t

or

y Adve

r

t

i

s

i

ng:

The

or

y a

nd Evi

de

nc

e i

n t

he

Pha

r

mac

e

ut

i

cal

I

ndus

t

r

y”by Wei Tan, State University of New York at Stony Brook

iv.

“

Ent

r

y a

nd Ext

e

r

na

l

i

t

y:

Hydr

oe

l

ec

t

r

i

c Ge

ne

r

a

t

or

s

i

n Br

a

zi

l

”by Rodrigo

Moita, IBMEC SAO PAULO

2. Search, Matching, and Investment

i.

“

Di

ve

r

s

i

t

y a

nd de

ma

nd e

xt

e

r

na

l

i

t

i

e

s

:

How c

heap i

nf

or

ma

t

i

on c

a

n r

e

duce

we

l

f

a

r

e

”by Heski Bar-Isaac, NYU

ii.

“

Too Ma

ny Bar

ga

i

ner

s

Spoi

l

The

Br

ot

h: The Impact of Bargaining on

Ma

r

ke

t

s

wi

t

h

Pr

i

c

e

Ta

ke

r

s

”by John Thanassoulis, Oxford University

iii.

“

Pr

i

c

e

Fi

xi

ng

a

nd

Non-Pr

i

c

e

Compe

t

i

t

i

on”by Marco Haan, University of

Groningen

iv.

“

Non-Coope

r

a

t

i

ve

Founda

t

i

ons

of

He

doni

c

Equi

l

i

br

i

um”by Michael Peters,

University of British Columbia

3. Auction Performance

i.

“

St

r

a

t

egi

c

Founda

t

i

ons

of

Pr

e

di

c

t

i

on Ma

r

ke

t

s

a

nd t

he

Ef

f

i

c

i

e

nt

Ma

r

ke

t

s

Hypot

he

s

i

s

”by Ricardo Serrano-Padial, University of California, San Diego

ii.

“

The

e

Bay Ma

r

ke

t

a

s

Se

que

nt

i

a

l

Se

c

ond Pr

i

ce

Auc

t

i

ons

---Theory and

Expe

r

i

me

nt

s

”by Joseph Wang, Caltech

iii.

“

Whe

n a

nd How t

o Di

s

ma

nt

l

e

Nuc

l

e

a

r

Wea

pons

:

Auc

t

i

on De

s

i

gn wi

t

h

Ext

e

r

na

l

i

t

i

e

s

”by Jingfeng Lu, National University of Singapore

iv.

“

Compe

t

i

ng

Auc

t

i

ons

wi

t

h

Endoge

nous

Qua

nt

i

t

i

es

”by Xianwen Shi, Yale

University

4. Partially Identified Models

i.

“

As

ympt

ot

i

c

be

ha

vi

or

of

i

ne

qua

l

i

t

y

c

ons

t

r

a

i

ne

d

mode

l

s

- with applications

t

o

of

f

i

c

e

s

uppl

y

s

upe

r

s

t

or

e

s

”by Arie Beresteanu, Duke University

ii.

“

We

a

k

I

de

nt

i

f

i

c

a

t

i

on

a

nd

Condi

t

i

ona

l

Mome

nt

Re

s

t

r

i

c

t

i

ons

”by Sung Jae

Jun, the Pennsylvania State University

iii.

“

Mor

e

on

Conf

i

de

nce

I

nt

e

r

va

l

s

f

or

Pa

r

t

i

a

l

l

y

I

de

nt

i

f

i

e

d

Pa

r

a

me

t

er

s

”by Joerg

Stoye, New York University

iv.

“

A

Si

mpl

e

Way

t

o

Ca

l

cul

a

t

e

Conf

i

de

nc

e

I

nt

er

val

s

f

or

Pa

r

t

i

a

l

l

y

I

de

nt

i

f

i

ed

Pa

r

a

me

t

e

r

s

”

by

Ti

e

me

n

Wout

e

r

s

e

n,

J

HU

5. Empirical Studies of Contracting and Asymmetric Information

i.

“

As

s

e

t

Li

qui

di

t

y

,

Bounda

r

i

e

s

of

t

he

Fi

r

m

a

nd

Fi

na

nc

i

a

l

Cont

r

a

c

t

s

:

Evi

de

nc

e

f

r

om

Ai

r

c

r

af

t

Le

a

s

e

s

”by Alessandro Gavazza, Yale University

ii.

“I

de

nt

i

f

yi

ng

a

nd

Te

s

t

i

ng

Mode

l

s

of

Hi

dde

n

I

nf

or

ma

t

i

on

a

nd

Mor

a

l

Ha

za

r

d:

An application to Man

a

ge

r

i

a

l

Compe

ns

a

t

i

on” by George-Levi Gayle,

Carnegie Mellon University

iii.

“

Se

mi

pa

r

a

me

t

r

i

c I

de

nt

i

f

i

c

a

t

i

on of

Mul

t

i

di

me

ns

i

ona

l

Sc

r

e

e

ni

ng Mode

l

s

”

by Heleno Pioner, The University of Chicago

iv.

“

Do La

wyer

s

Wor

k f

or

Cl

i

e

nt

s

? Us

i

ng Ti

mi

ng of

Dr

oppe

d Di

s

put

e

s

t

o

Estima

t

e

I

nc

e

nt

i

ve

Mi

s

al

i

gnme

nt

”by: Yasutora Watanabe, Northwestern

University (KSM)

6. Contract Theory

i.

“

Hol

d-up a

nd Dur

a

bl

e

Tr

a

di

ng Oppor

t

uni

t

i

e

s

”

by Chr

i

s

t

opher

Wi

gna

l

l

,

UCSD

ii.

“

Ri

gi

di

t

y i

n Nonl

i

nea

r

Pr

i

c

i

ng unde

r

Hi

dde

n I

nve

s

t

me

nt

”by Rui Zhao,

University at Albany, SUNY

University of Zurich

iv.

“

Whe

n

Shoul

d

Cont

r

ol

Be

Sha

r

e

d?”

by

Pa

ul

Mi

l

gr

om,

St

a

nf

or

d

Uni

ve

r

s

i

t

y

7. Demand Estimation

i.

“

Doe

s

Buyi

ng

Hi

gh

Me

an

Buyi

ng

Of

t

e

n?

Qua

l

i

t

y

Choi

c

e

a

nd

Re

pl

a

ce

me

nt

Cycl

e

s

f

or

a

Ve

r

t

i

c

a

l

l

y

Di

f

f

e

r

e

nt

i

a

t

e

d,

Ra

pi

dl

y

I

mpr

ovi

ng

Dur

a

bl

e

Good”by

Jeffrey Prince, Cornell University

ii.

“

De

r

i

vi

ng

t

he

Opt

i

ma

l

Pr

i

c

i

ng

St

r

a

t

egy

Gi

ve

n

t

he

Ef

f

ec

t

of

De

a

l

i

ng

Pa

t

t

e

r

n

on Cons

ume

r

Be

ha

vi

or

:

A St

r

uc

t

ur

a

l

Appr

oac

h”

by Pr

i

s

c

i

l

l

a

Me

de

i

r

os

,

Northwestern University

iii.

“I

nt

r

a

-Ho

us

e

hol

d

Cons

umpt

i

on

Ext

e

r

na

l

i

t

i

e

s

of

Ce

l

l

ul

a

r

Phone

Se

r

vi

c

e

”

by

Ching-I Huang, National Taiwan University

iv.

“

A

Hybr

i

d

Di

s

c

r

e

t

e

Choi

c

e

Mode

l

of

Di

f

f

e

r

e

nt

i

at

e

d

Pr

oduc

t

De

ma

nd

wi

t

h

a

n Appl

i

c

a

t

i

on t

o Pe

r

s

ona

l

Comput

e

r

s

”by Minjae Song, University of

Rochester

8. Topics in Applied Microeconomics

i.

“

Ai

r

Qua

l

i

t

y

a

nd

Ne

i

ghbor

s

'

Ef

f

ec

t

s

:

Evi

de

nc

e

f

r

om

Sur

vi

va

l

Ana

l

ys

i

s

”by

Mariano Rabassa, University of Illinois Urbana-Champaign

ii.

“

Do

Ha

r

s

her

Puni

s

hme

nt

s

De

t

e

r

Cr

i

me

?

Pe

r

c

e

pt

i

ons

a

nd

Be

ha

vi

or

Ar

ound

t

he

Age

of

Cr

i

mi

na

l

Ma

j

or

i

t

y”by Randi Hjalmarsson, University of

Maryland

iii.

“

The

I

mpac

t

of

Poor

He

a

l

t

h

on

Educ

a

t

i

on:

New

Evi

de

nc

e

Us

i

ng

Ge

net

i

c

Ma

r

ke

r

s

”

by

We

i

l

i

Di

ng,

Que

e

n'

s

Uni

ver

s

i

t

y

iv.

“

Pol

i

t

i

c

a

l

Pr

i

c

e

Cyc

l

e

s

i

n

Re

gul

a

t

e

d

I

ndus

t

r

i

e

s

:

The

or

y

a

nd

Evi

de

nc

e”by

Rodrigo Moita, IBMEC SAO PAULO

9. Dynamics and Econometrics in IO

I

ndus

t

r

y”by Allan Collard-Wexler, New York University

ii.

“

Nonpa

r

a

me

t

r

i

c

I

de

nt

i

f

i

ca

t

i

on

a

nd

Es

t

i

ma

t

i

on

of

Fi

ni

t

e

Mi

xt

ur

e

Models of

Dyna

mi

c

Di

s

c

r

e

t

e

Choi

c

e

s

”

by:

Hi

r

oyuki

Ka

s

a

har

a

,

Uni

ve

r

s

i

t

y

of

We

s

t

e

r

n

Ontario

iii.

“

An

Appr

oa

c

h

t

o

I

de

nt

i

f

i

c

a

t

i

on

of

Mar

gi

na

l

Ef

f

e

c

t

s

i

n

a

Cor

r

e

l

a

t

e

d

Ra

ndom

Ef

f

ec

t

s

Mode

l

f

or

Pa

ne

l

Da

t

a

”

by

Chr

i

s

t

i

a

n

Ha

ns

e

n,

Uni

ve

r

s

i

t

y

of

Chi

c

ago

10. Auctions, Betting, and Risk Aversion

i.

“

Ra

t

i

ona

l

Expe

c

t

a

t

i

ons

a

t

t

he

Ra

c

e

t

r

ac

k

:

Te

s

t

i

ng

Expe

c

t

e

d

Ut

i

l

i

t

y

The

or

y

Us

i

ng

Be

t

t

i

ng

Ma

r

ke

t

s

”by Amit Gandhi, University of Chicago

ii.

“

Engl

i

s

h Auc

t

i

ons

wi

t

h Re

s

a

l

e

:

An Expe

r

i

ment

a

l

St

udy”by Sotirios

Georganas, Bonn Graduate School of Economics

iii.

“

St

r

uc

t

ur

a

l

i

nf

e

r

e

nc

e

s

f

r

om Fi

r

s

t

-Pr

i

c

e

Auc

t

i

on da

t

a

”by Andres Romeu,

Universidad de Murcia

iv.

“

De

c

i

s

i

on

Ma

ki

ng

Under

Ri

s

k

i

n

De

a

l

or

no

De

a

l

”by Nicolas de Roos,

University of Sydney

11. Labor Economics Topics

i.

“

The

Ef

f

e

c

t

s

of

Shor

t

-Term

Training

Measures

on

Individual

Une

mpl

oyme

nt

Dur

a

t

i

on

i

n

We

s

t

e

r

n

Ge

r

ma

ny”by Stephan Thomsen, ZEW,

Mannheim

ii.

“

Coa

s

e

Wi

ns

Af

t

e

r

Al

l

:

No-Fault v.s. Unilateral Divorce Laws and Divorce

Ra

t

e

s

”by Marjorie McElroy, Duke University

iii.

“

Di

d

We

l

f

a

r

e

Ref

or

m

I

mpr

ove

t

he

Ac

a

de

mi

c

Pe

r

f

or

mance of Children in

Low-I

nc

ome

Hous

e

hol

ds

?”

by

Le

i

Zha

ng,

Cl

e

ms

on

Uni

ve

r

s

i

t

y

iv.

“

How

We

l

l

a

r

e

Ea

r

ni

ngs

Me

a

s

ur

e

d

i

n

t

he

Cur

r

e

nt

Popul

a

t

i

on

Sur

vey?

Bi

a

s

f

r

om

Nonr

e

s

pons

e

,

I

mput

a

t

i

on,

a

nd

Pr

oxy

Re

s

ponde

nt

s

”

by

Ba

r

r

y

Hi

r

s

ch,

Trinity University

我個人的文章，則是在 6 月 23 日早上第一場的 session 發表，該場次的四個發表

人當中，竟然有三個同樣是與 Northwestern University 有關連的，真是有趣的巧

合。在我發表之後，Yasutora Watanabe、Tao-Yi Wang、Hiroshi Ohashi 等人，都

給了我相當具體的改進建議。

除此之在，大會也邀請了許多著名學者來發表最新的研究成果。我參加的這些

plenary sessions ， 聽 到 以 下 這 些學 者 的報告 : Robert Miller, Kenneth Wilpin,

Jonathan Levin, Patrick Bajari, C. Lanier Benkard。他們當中有些人的文章，正好是

我最近正在研讀的，能有機會聽到作者親自報告，對於瞭解文章的概念相當有幫

助。

在研討會結束，我飛到了芝加哥，到 Northwestern University 進行訪問。主要是

跟 Christopher Taber 和 Robert Porter 兩位教授進行討論。我跟 Christopher Taber

有一篇共同的 working paper，受限於地理上的距離跟時差的關係，一直很難進行

充分的討論，透過這次的面會面溝通，我們解決了一些文章當中的難題，期望在

近期了就能夠將文章完成。Robert Porter 則是針對於一些我目前持續在進行的研

究，提供了我很多具體的建議。也給了我將文章投稿至期刊的意見，相信會對於

這些文章的投稿有很大的幫助。

The paper I discussed with Professor Christopher Taber is about estimating the peer

effects of a universal voucher program in high schools. Such a program will

encourage more students to attend private schools. While private schools presumably

have better quality, the policy may hurt students staying in public schools after the

implementation of the policy. This is because the students who may transfer to private

schools are likely to be those with better demographic background. Therefore, the

average quality of the students staying in public schools becomes worse. The worsen

quality may, in turn, hurts those staying in public schools.

To estimate the potential policy impacts, we estimate a model with school choice and

student outcome. In the meetings during my visit, we were trying to figure out the

correct formulation to account for the unobservables, both at the individual level and

the school level.

As for the meetings with Professor Robert Porter, I was seeking for his comments for

several of my working papers, including estimating the demand for cellular phone

service using a BLP-style method, estimating the demand with intra-household effects,

and a model to explain intra-network discounts in the cellular phone industry. He

suggested me a couple of feasible approaches, which I tried to apply to my researches

during my visit in Chicago. Moreover, he gives some guidance as for where to submit

my papers. Hopefully, his experiences about top economics journals will help me to

submit my works to appropriate journals and have them published soon.

In addition, I also invite Professor Robert Porter to visit Taiwan in the future. It seems

he is interesting in doing so as long as he has free time. Nonetheless, I am not sure

whether the Economics Department in NTU will have sufficient budget for his

visiting. Therefore, anything about his visiting is still uncertain.