2.1 Setting

2.1.2 Election

We consider an election in which a winner is decided by a plurality rule, with ties being settled using an equal-probability rule. If only one candidate runs for the election, this candidate wins with certainty. In this election, voters vote sincerely. Thus, they play only weakly undominated strategies. We analyze voters’ behavior by extending the probabilistic voting model of Persson and Tabellini (Reference Persson and Tabellini2000).

The voters’ payoff from a candidate depends on two items: the size of policy g and the type of candidates. Voter i's payoff is defined as (1 − λ_D)d_D + ω_i when a dynastic candidate wins, and as (1 − λ_N)d_N when a non-dynastic candidate wins. The variable ω_i represents voter i’s preference for a type of candidate (i.e., dynastic or non-dynastic) and the valence of candidates (e.g., personality and party affiliation). If (1 − λ_D)d_D = (1 − λ_N)d_N, voter i prefers a dynastic candidate when ω_i > 0, and prefers a non-dynastic candidate when ω_i < 0. We assume that ω_i is uniformly distributed on [γ − 1/(2ϕ), γ + 1/(2ϕ)] where ϕ represents the degree of preference homogeneity within the district. The random variable γ represents an aggregate popularity shock and is uniformly distributed on [ − 1/(2ξ), 1/(2ξ)]. It is realized after the candidates announce their policy platforms. That is, candidates are uncertain about the change of voters’ preference between platform choice and the election. Voter i votes for a dynastic candidate whenever ω_i + (1 − λ_D)d_D − (1 − λ_N)d_N > 0. Then, we define the (random) share of voters who support a dynastic candidate as ϕ[(1 − λ_D)d_D − (1 − λ_N)d_N + γ + 1/(2ϕ)].

Given the uniform distribution of γ, the probability that a dynastic candidate wins, denoted as π(λ_D, λ_N), is

(1)\begin{eqnarray*} &&\pi \left( {\lambda _D ,\lambda _N } \right) \\ &&\quad= \left\{\begin{array}{c@{\hspace*{7pt}}l} 1 & {{\rm if}\xi \left[ {\left( {1 - \lambda _D } \right)d_D - \left( {1 - \lambda _N } \right)d_N } \right] > \frac{1}{2}} \\ 0 & {{\rm if}\xi \left[ {\left( {1 - \lambda _D } \right)d_D - \left( {1 - \lambda _N } \right)d_N } \right] < - \frac{1}{2}} \\ {\frac{1}{2} + \xi \left[ {\left( {1 - \lambda _D } \right)d_D - \left( {1 - \lambda _N } \right)d_N } \right].} & {} \\ \end{array} \right.\\ \end{eqnarray*}

Otherwise, because the condition for the victory of a dynastic candidate, assuming an interior solution, is ϕ[(1 − λ_D)d_D − (1 − λ_N)d_N + γ + 1/(2ϕ)] > 1/2. This is rewritten as γ > −[(1 − λ_D)d_D − (1 − λ_N)d_N]. If only one candidate runs, voters vote on this candidate, and this candidate wins with certainty.

We now consider the decision candidates make on whether to run. Candidates are office motivated and rent-seeking. The benefit from holding office is v > 0, which is not related to a policy. Candidates also benefit from a rent policy, r = λ_jd_j. We assume that candidates can obtain r if they win. Note that even if candidates obtain a positive benefit that is not the same as r, the result does not change as long as they obtain a part of r. Candidates announce and commit to the amount of distributions that will be allocated to the growth policy, 1 − λ, before voting. Denote V_j as the expected benefit of a type-j candidate. Then, the respective expected benefits of dynastic and non-dynastic candidates are

(2)\begin{equation} V_D \equiv \pi \left( {\lambda _D ,\lambda _N } \right)\left( {\lambda _D d_D + v} \right), \end{equation}

(3)\begin{eqnarray} V_N \equiv \left[ {1 - \pi \left( {\lambda _D ,\lambda _N } \right)} \right]\left( {\lambda _N d_N + v} \right).\end{eqnarray}

If dynastic and non-dynastic candidates decide to run, they have to pay the cost of running, c_D and c_N, respectively. We assume that the cost is higher for non-dynastic candidates than for dynastic candidates (i.e., c_D < c_N). This assumption is drawn from some observable features of dynastic candidates. First, dynastic candidates can inherit a well-organized support group from their relative. The support group helps candidates mobilize voters in elections. Second, dynastic candidates can inherit financial resources from their relative. Third, constituents easily recognize the name of dynastic candidates (Feinstein, Reference Feinstein2010).Footnote ⁹ These advantages decrease the cost of running for dynastic candidates, but increase the cost of running for competing non-dynastic candidates. We call this assumption the electoral advantage.Footnote ¹⁰

In addition, we assume the following:

Assumption 1c_D < 1/(4ξ).

This assumption means that a dynastic candidate decides to run even if this candidate has no bargaining advantage (α = 1).Footnote ¹¹ If this assumption is dropped, there may exist an unrealistic equilibrium in which no one runs.

2.2 Equilibrium

2.2.1 Legislative bargaining

We begin by analyzing legislative bargaining. As mentioned previously, we consider an ultimatum game: if a proposal is not approved, the game ends, and each member receives $\skew3\bar{d}$. This means that the agenda setter proposes $\skew3\bar{d}$ to each member of the majority coalition, and keeping all the remaining resources ($1 - \left( {{\rm L} - 1} \right)\skew3\bar{d}/2 > \skew3\bar{d}$), which the other members approve (see Proposition 1 of Baron and Ferejohn (Reference Baron and Ferejohn1989)). Thus, legislators who are not in the majority coalition cannot get anything (d_l = 0) if a proposal is approved. Accordingly, the expected values of the distribution for the district are

\begin{eqnarray*} d_N &\equiv& \left( {\frac{1}{{L + \left( {\alpha - 1} \right)n}}} \right)\left( {1 - \frac{{L - 1}}{2}\skew3\bar{d}} \right) + \left( {1 - \frac{1}{{L + \left( {\alpha - 1} \right)n}}} \right)\frac{{\skew3\bar{d}}}{2},\\ d_D & \equiv & \left( {\frac{\alpha }{{L + \left( {\alpha - 1} \right)\left( {n + 1} \right)}}} \right)\left( {1 - \frac{{L - 1}}{2}\skew3\bar{d}} \right) + \left( {1 - \frac{\alpha }{{L + \left( {\alpha - 1} \right)\left( {n + 1} \right)}}} \right)\frac{{\skew3\bar{d}}}{2}. \end{eqnarray*}

The first term is the expected payoff to be an agenda setter, and the second term is the expected payoff to be a member of the majority coalition. The district with a dynastic legislator is expected to obtain more benefits than the one with a non-dynastic legislator. That is, d_D > d_N. As α increases, d_D increases and d_N decreases. In other words, the larger bargaining power of dynastic legislators results in a larger allocation of benefits to their districts.

2.2.2 Campaign platform and election returns

Given the predicted outcomes of legislative bargaining, we next consider the size of λ in the campaign platform and the expected vote share. To ensure interior solutions for λ_j ∈ (0, 1) for j = D, N, we assume that

Assumption 2 (d_D − d_N)/3 < 1/(2ξ) − v < (d_D + 2d_N)/3.

In other words, the value of ξ is not too high or too low. Without this assumption, λ_j = 0 or 1 in equilibrium.Footnote ¹²

By differentiating V_j in (2) and (3) with respect to λ_j, we can derive the following values of λ*_j announced by the candidates as a campaign platform in equilibrium

(4)\begin{equation} \lambda _D^* \equiv \frac{1}{3} - \frac{1}{{d_D }}\left( {v - \frac{1}{{2\xi }} + \frac{1}{3}d_N } \right),\end{equation}

(5)\begin{equation} \lambda _N^* \equiv \frac{1}{3} - \frac{1}{{d_N }}\left( {v - \frac{1}{{2\xi }} + \frac{1}{3}d_D } \right).\end{equation}

After substituting π(λ_D, λ_N) for λ*_D and λ*_N in the third line of (1), the probability that a dynastic candidate wins in equilibrium becomes

(6)\begin{equation} \pi ^* \equiv \frac{1}{2} + \frac{\xi }{3}\left( {d_D - d_N } \right).\end{equation}

Given Assumption 2, π* ∈ (0, 1). The expected vote share of a dynastic candidate is

(7)\begin{equation} \frac{1}{2} + \frac{\varphi }{3}\left( {d_D - d_N } \right)\end{equation}

since the mean of γ is 0.

Next, substitute (4), (5), and (6) into V_j in (2) and (3). Then, the expected benefits of dynastic and non-dynastic candidates in equilibrium are

(8)\begin{equation} V_D^* \equiv \left[ {\frac{1}{{2\xi }} + \frac{1}{3}\left( {d_D - d_N } \right)} \right]^2 ,\end{equation}

(9)\begin{equation} V_N^* \equiv \left[ {\frac{1}{{2\xi }} - \frac{1}{3}\left( {d_D - d_N } \right)} \right]^2 .\end{equation}

Thus, the ratio of rent, the probability of winning, the expected vote share, and the expected benefit are higher for a dynastic candidate than for a non-dynastic candidate because d_D > d_N.

2.3 Theoretical implications

We can draw three empirical predictions from our model. First, our model of legislative bargaining predicts the following hypothesis:

H1Dynastic legislators bring more distributions to their districts than do non-dynastic legislators.

Second, the model leads to predicting an election outcome. The probability that a dynastic candidate wins in a competitive district is given by (6), which is always higher than one half. The expected vote share is given by (7), which is also higher than one half. In contrast, a non-dynastic candidate's probability of winning in a competitive district is lower than one half. In the dynastic district, a dynastic candidate will win with certainty. Drawing on these results, we expect that:

H2Dynastic candidates enjoy a higher probability of winning and a higher vote share than non-dynastic candidates do.

Third, we examine the economic consequences of electing dynastic and non-dynastic candidates. As mentioned previously, we interpret g ( = (1 − λ_j)d_j) as a growth policy aimed at improving the socioeconomic infrastructure and expanding job opportunities within the district, leading to higher economic performance. Electing dynastic legislators should result in higher economic performance than electing non-dynastic legislators because dynastic legislators bring more allocations (i.e., the larger amount of d_j) to their districts. Importantly, this relationship holds only in a competitive district, where dynastic legislators generate a higher expected amount of spending on policy g than non-dynastic legislators do (i.e., (1 − λ*_D)d_D > (1 − λ*_N)d_N).Footnote ¹⁴ On the other hand, electing dynastic legislators in the dynastic district results in lower economic performance than when electing non-dynastic legislators in the competitive district. This is because those dynastic legislators set λ_D = 1 and win with certainty. In other words, dynastic legislators from the dynastic district only spend the distributive benefits on the limited segment of the electorate, including themselves, in the district, which makes no contribution to the economic performance of the entire district. Taken together, we obtain the following corollary:

Corollary 1In the competitive district, economic performance is higher when a dynastic legislator is elected than when a non-dynastic legislator is elected. (ii) Economic performance is higher in the competitive district with a non-dynastic legislator than in the dynastic district (with a dynastic legislator).

For our empirical analysis, we have to specify whether dynastic legislators, if elected, are from the dynastic or competitive district. According to Proposition 1, the type of district is determined by V*_N and c_N. We focus on the role of c_N, arguing that the dynastic district appears more likely when c_N is sufficiently high, while the competitive district appears more likely when c_N is low.Footnote ¹⁵ The value of c_N is strongly influenced by the amount of electoral resources dynastic candidates possess. This is because a non-dynastic candidate has to spend more on mobilization to challenge the dynastic opponent, who has the larger amount of inherited resources. More resource-rich dynastic candidates increase the campaign costs for non-dynastic candidates, discouraging non-dynastic candidates from running. Note that dynastic candidates are, on average, more resource-rich than non-dynastic candidates, because dynastic candidates can inherit a well-organized support group, financial resources, and name recognition among the constituents from their relative. Thus, a larger gap in electoral resources makes it particularly challenging for non-dynastic candidates to fight against dynastic candidates. In contrast, less-resourceful dynastic candidates lower the campaign cost for non-dynastic candidates.

In short, we predict that the dynastic district appears when dynastic candidates are more resourceful, while the competitive district appears when dynastic candidates are less resourceful. Accordingly, we develop the following hypothesis:

H3Compared to the districts electing non-dynastic legislators, the districts electing dynastic legislators with abundant resources display lower economic performance, whereas the districts electing dynastic legislators without abundant resources display higher economic performance.

3.1 Fiscal transfers

We test the first hypothesis by examining the flow of intergovernmental transfers to municipalities. We use panel data that include biographical and electoral information of the representatives elected for the Lower House in 1996, 2000, 2003, and 2005, and annual fiscal and socioeconomic information of municipalities between 1997 and 2007.Footnote ¹⁷ We also present an instrument variable estimation in order to address concerns regarding endogeneity.

We estimate the following model

(10)\begin{eqnarray} [Transfer]_{ml,t} & = &\beta [dynasty]_{ml,t - 1} + \gamma _1 [LDP]_{ml,t - 1} + \gamma _2 [Term]_{ml,t - 1} \\ && + \,\gamma _3 [M\arg in]_{ml,t - 1} + {\rm \lambda }{\bf w}_{ml,t} + \phi _t + \rho _{ml} + \in _{ml,t} ,\end{eqnarray}

where [Transfer]_{ml, t} denotes the log of the amount of fiscal transfers from the national government to municipality m in district l in year t. The amount of government transfer is the sum of two broad categories of governmental transfers, namely the local allocation tax grant and the national treasury disbursement.Footnote ¹⁸ [Transfer]_{ml, t} is a per capita amount and is measured in 1,000 yen. The average equals 5.270 with the standard error 0.918 for our study period.

[Dynasty]_{ml, t − 1} in (10) is our key explanatory variable and equals 1 if municipality m of district l in yeart − 1 is represented by a dynastic legislator and 0 otherwise. We define ‘dynastic’ legislators as those whose parents were members of the Lower House in the past. The percentages of dynastic legislators elected in each election are 28% in 1996, 25% in 2000, 27% in 2003, and 26% in 2006.Footnote ¹⁹ Similarly, [LDP]_{ml, t − 1} equals 1 if the municipality is represented by a legislator who belongs to the Liberal Democratic Party (LDP) and 0 otherwise. The LDP was in power during the period of our study. [Term]_{ml, t − 1} denotes the number of previous terms of the legislator representing the municipality, while [Margin]_{ml, t − 1} denotes the margin of victory.

The vector w_{ml, t} includes the characteristics of municipalities, such as per capita income, the fiscal strength index, the total size of population, the proportion of population under 15 years old and over 65 years old, the proportion of population in the agricultural sector and service sector, the degree of urbanization, and population density.Footnote ²⁰ The financial strength index measures the ratio of municipal financial revenue to financial demand. When the revenue is exactly equal to the anticipated expenditure, the index becomes 1. The scores below 1 indicate that the demand surpasses the revenue, while the scores above 1 indicate that the revenue surpasses the demand. The degree of urbanization is measured by the ratio of population living in Density Inhabited Districts (DID). ρ_ml and φ_t are municipality and year fixed effects, respectively. ε_{ml, t} is a municipality-year specific error term.Footnote ²¹ The number of municipalities in Japan was slightly more than 3,300 until about 2003, but since then the number has dropped to around 1,800 due to large-scale mergers that happened throughout Japan. In total, our municipality-level data include about 30,000 observations for 11 years. The summary statistics of the variables for our regression analysis are reported in Table 1.

Table 1.

Summary statistics for municipality-level data

Table 2 reports the estimation results. The standard errors are clustered by municipalities. Column (1) in Table 2 shows that municipalities receive more transfers when they are represented by dynastic legislators as compared to when non-dynastic legislators represent them. The difference is estimated to be statistically significant. The coefficient associated with the dynastic legislator dummy in column (1) indicates that municipalities expect a 1.5% increase in the transfer when they are represented by a dynastic legislator. This result offers support for the first hypothesis.

Table 2.

The effect of dynastic legislators on fiscal transfers to municipalities

Notes: Table entries are regression coefficients with robust standard errors in parentheses. Standard errors are clustered by municipalities. The estimates are based on the municipality-level data from 1997 to 2007. The dependent variable is the total amount of fiscal transfer per capita from the national government to municipalities. The dependent variable is logged for estimation. The municipality and year fixed effects are included in the models. ^∗∗ p < 0.05 and ^∗ p < 0.10.

This suggests that municipalities receive more transfers from the national government when they are represented by dynastic legislators. However, one may argue that the relationship is spurious. For example, suppose that an area with a deteriorating economic situation elects a dynastic legislator because it expects him to secure more grants to improve its economic condition. Further, this area may receive more transfers from the government simply because it has greater financial needs. If this is the case, the effect of dynastic legislators is overestimated. Alternatively, one may argue that there is a reverse causality (i.e., our dependent variable determines the presence of dynastic legislators) because an increase in subsidies enhances the chance of reelecting dynastic legislators who have brought back the money to the district. Finally, suppose that there is an unobservable ability that helps politicians secure more funds to their constituencies. If dynastic politicians tend to have such a trait, our results are likely to be biased because we are unable to include it in the estimation.

We address these potential concerns for endogeneity by using the gender of the previous incumbent's children as an instrumental variable.Footnote ²² Specifically, we use the fraction of boys among the predecessor's children. Given that more than 90% of Japanese politicians between 1996 and 2007 were males, if a politician has only daughters, he is unlikely to give his seat to one of his children. In fact, among the 124 politicians coded as dynastic in our dataset, only three are female.Footnote ²³ Thus, the fraction of male children of a politician should be highly and positively correlated with the chance of dynastic inheritance of the politician's seat. However, the gender of offsprings (of the previous incumbent) is unlikely to be related to the amount of transfers in the current period. These two properties make it an ideal instrument.

Let [Boy]_{ml, prev} be the fraction of male children among all children of the previous incumbent in district l. We calculate [Boy]_{ml, prev} as follows. First, for each politician in our dataset, we checked if the member who served before him or her belonged to the same party. If not, [Boy]_{ml, prev} is automatically coded as 0, because the dynastic inheritance of seats cannot happen.Footnote ²⁴ Second, if the previous incumbent was from the same party, then we consulted a book called ‘Jinji Koshin Roku’ that lists the family information of notable people in Japan.Footnote ²⁵

Then, we run our first-stage regression as follows

(11)\begin{equation} \left[ {Dynasty} \right]_{ml,t - 1} = \beta \left[ {Boy} \right]_{ml,prev} + \lambda {\bf X}_{ml,t} + \phi _t + \rho _{ml} + \varepsilon _{ml,t} ,\end{equation}

where X_{ml, t} contains all other independent (i.e., exogenous) variables included in (10). The first-stage regression produces a statistically significant coefficient on [Boy]_{ml, prev} ($\hat \beta $ = 0.268, SE = 0.046), suggesting that it is a valid strong instrument.Footnote ²⁶

We then run our second-stage regression using a fitted value of [Dynasty]_{ml, t − 1} in (11). The result is reported in column (2) in Table 2. The coefficient associated with [Dynasty]_{ml, t − 1} is positive and statistically significant. The results do not change substantially even if we use the number of male children as an alternative instrument. The estimation results from IV regressions suggest that the presence of dynastic legislators has a strong causal impact on the allocation of distributive benefits to municipalities. In short, our instrumental variable estimation also lends support to the hypothesis that dynastic legislators bring more distributive benefits to the district than non-dynastic legislators.

3.2 Election returns

The second hypothesis predicts that dynastic candidates have a higher probability of winning and a higher vote share as compared to non-dynastic candidates. We test this hypothesis by estimating the following model

(12) \begin{eqnarray} [{\rm Re} turn]_{h,l} &=& \beta [dynasty]_{h,l} + \gamma _1 [Female]_{h,l} + \gamma _2 [Age]_{h,l} + \gamma _3 [Local]_{h,l} \\ &&+\, \gamma _4 [LDP]_{h,l} + \gamma _5 [DP]_{h,l} + \gamma _6 [Komei]_{h,l} \\ &&+\, \gamma _7 [Incumbent]_{h,l} + \gamma _8 [Term]_{h,l} + \rho _l + \in _{h,l} , \end{eqnarray}

where [Return]_{h, l} is either a dichotomous variable that equals 1 if candidate h wins an election in district l or a continuous variable that equals the vote shares of candidate h in district l. For the dichotomous variable, candidates are assigned 0 even if they gained a seat in the PR tier after being defeated in a single-member district. The vote share is equal to the number of candidate's votes divided by the total number of eligible votes cast, multiplied by 100. ρ_l denotes a district-fixed effect, while ε_{h, l} denotes a candidate-specific error term. With ρ_l, the model exploits the variation across candidates within a district.

[Dynasty]_{h, l} in (12) equals 1 if candidate h in district l is dynastic. Because it is extremely challenging to find the dynastic status of those who lost elections, we limit our search to candidates who ran in the 2005 General Election (N = 989). Thus, in this part of the analysis, the sample represents candidates who ran in 2005. The percentage of dynastic candidates is 11% in the 2005 election. In addition to [Dynasty]_{h, l}, equation (12) includes several variables that control the effects of the demographic and political attributes of the candidates on election returns. [Female]_{h, l} equals 1 if a candidate is female. [Age]_{h, l} denotes the age of the candidates in 2005. [Local]_{h, l} equals 1 if a candidate served as a governor, mayor, or representative at the prefecture or municipality level before running for the Lower House election. [LDP]_{h, l}, [DPJ]_{h, l}, and [Komei]_{h, l} denote the party affiliation of the candidates. [LDP]_{h, l} equals 1 if a candidate is from LDP. [DPJ]_{h, l} equals 1 if a candidate belongs to the Democratic Party of Japan. [Komei]_{h, l} equals 1 if a candidate belongs to the Komei party.Footnote ²⁷ [Incumbent]_{h, l} is a dummy variable for the incumbency status of candidates. [Term]_{h, l} is the number of times that the candidate has been elected to the Lower House. The summary statistics are reported in Table 3.

Table 3.

Summary statistics for candidate-level data

Table 4 presents the estimation results. We estimate a logit model when the dependent variable is the dichotomous variable of winning. The standard errors are clustered by districts. According to column (1), the positive and statistically significant coefficient associated with [Dynasty]_{h, l} suggests that dynastic candidates are more likely to win the 2005 election as compared to non-dynastic counterparts. The estimated coefficient in column (1) suggests that their probability of winning is 22% higher than that of non-dynastic candidates.Footnote ²⁸ When the vote share is the dependent variable, we estimate the standard fixed effect model. Column (2) of Table 4 suggests that the coefficient associated with the dynastic status is positive and statistically significant. That is, dynastic candidates enjoy higher vote shares as compared to non-dynastic candidates. The estimated coefficient in column (2) implies that the average vote share of dynastic candidates is 5.273 percentage points higher than that of non-dynastic candidates. These results show evidence in support for the second hypothesis.

Table 4.

The effect of dynastic candidates on the election returns in 2005

Notes: Table entries are logit regression coefficients in Column (1) and linear regression coefficients in Column (2) with robust standard errors in parentheses. Standard errors are clustered by districts. The estimates are based on candidates running for the 2005 Lower House election. The dependent variables are a binary variable that equals 1 if a candidate won the election in column (1), and the percent of vote share in column (2). The district fixed effects are included in the models. ^∗∗ p < 0.05 and ^∗ p < 0.10.

3.3 Economic performance

To test the hypothesis on economic performance (H3), we determine whether dynastic legislators are resource-rich or not by using the data on the reported amount of financial resources. Specifically, we rely on the total amount of family assets of dynastic legislators. We assume that more family assets allow dynastic candidates to spend more on campaign activities, which results in a higher cost of campaign for non-dynastic candidates. The data on the assets are available for 1996, 2000, 2003, and 2005. We code dynastic legislators as rich if their family assets exceed the mean of the reported assets among all legislators, and poor otherwise.Footnote ²⁹ Our regression analysis reports that dynastic candidates own more family assets than non-dynastic candidates. Column (1) in Table 5 reports that the difference in the family assets is about 5 million yen, after controlling for several political and demographic attributes. Column (2) in Table 5 shows a similar result wherein dynastic legislators are more resource-rich than non-dynastic legislators when the total amount of campaign revenue is used as a measure of resources.Footnote ³⁰

Table 5.

Dynastic legislators and financial resources

Notes: Table entries are linear regression coefficients with robust standard errors in parentheses. Standard errors are clustered by prefectures. The estimates are based on winners elected for the 1996, 2000, 2003, and 2005 Lower House elections in Column (1), and in the 1996 election in Column (2). The dependent variable in column (1) is the total asset in 1,000,000 yen, while the dependent variable in column (2) is the total campaign revenue in 1,000,000 yen. The prefecture fixed effects are included in the models. ^∗∗ p < 0.05 and ^∗ p < 0.10.

As a measure of local economic performance, we use the rate of GDP growth in prefectures. We expect that the prefectures represented by dynastic legislators achieve the lower levels of income and GDP growth than the prefectures represented by non-dynastic legislators. In this subsection, we use the prefecture-level data because the data on economic growth are unavailable at the municipality level.Footnote ³¹

We estimate the following model

(13)\begin{eqnarray} {[}Growth{]}_{k,t} &=& \alpha {[}Growth{]}_{k,t - 1} + \beta _1 {[}P.RichDynasty{]}_{k,t - 2} + \beta _2 {[}P.RichDynasty{]}_{k,t - 3} \\ &&+\, \beta _3 {[}P.PoorDynasty{]}_{k,t - 2} + \beta _4 {[}P.PoorDynasty{]}_{k,t - 3} \\ &&+\, \gamma _1 {[}P.LDP{]}_{k,t - 2} + \gamma {[}P.LDP{]}_{k,t - 3} \\ &&+\, \gamma _3 {[}Transfer{]}_{k,t - 1} + \gamma _4 {[}Transfer{]}_{k,t - 2} \\ && +\, \lambda {\bf w}_{k,t} + \phi_t + \rho_k + \in_{k,t} , \end{eqnarray}

where [Growth]_{k, t} denotes the annual rate of GDP growth in prefecture k in year t. The right-hand side of (13) includes the lagged rate of growth, the lagged proportions of dynastic and LDP legislators, socioeconomic variables, and fixed effects for prefectures and years. Following prior research on economic growth (Caselli et al., Reference Caselli, Esquivel and Lefort1996), we include the lagged dependent variable in the right-hand side. Further, we take the log of the growth rate.

[P.RichDynasty]_{k, t − 2} measures the proportion of dynastic legislators who were elected in prefecture k in year t − 2 and whose family assets are above the mean of the sample, while [P.PoorDynasty]_{k, t − 2} measures the proportion of dynastic legislators who were elected in prefecture k in year t − 2 whose family assets are below the mean. [P.RichDynasty]_{k, t − 3} and [P.PoorDynasty]_{k, t − 3} are similarly defined. We use lags at t − 2 and t − 3 because there should be a time lag before projects and government investment can affect local economic performance. Legislators at t determine the amount of transfers at t + 1, which then affects the rate of income and GDP growth at t + 2 or later. Further, these lags also minimize the possibility of the reverse causality between the presence of dynastic legislators and economic performance. In order to calculate the proportions of resource-rich and resource-poor dynastic legislators, we first code each member of the Diet for their dynastic status. Once a legislator's family background is coded, we simply sum the number of resource-rich and resource-poor dynastic legislators using their assets in prefecture k in year t − 2 (or t − 3) and divide it by the total number of lawmakers in prefecture k. We expect β ₁ and β ₂ to be negative and β ₃ and β ₄ to be positive. We also include [P.LDP]_{k, t − 2} and [P.LDP]_{k, t − 3} to capture the effect of belonging to the party in power. These variables are equal to the proportion of representatives from LDP and are defined in the same way as [P.RichDynasty] and [P.PoorDynasty].

Equation (13) includes a measure of fiscal transfers from the national government to prefectures. We use a log of the total amount of government transfers per capita at t − 1 and t − 2. We assume that fiscal transfers are expected to improve the GDP growth rate. Thus, the coefficients associated with [Transfer] should be estimated to be positive. We use the lags of [Transfer] at t − 1 and t − 2 because fiscal transfers are not likely to have an immediate impact on local economic performance.

Additionally, we include w_{k, t} and ρ_k in order to control the effects of underlying socioeconomic characteristics of prefecture k on [Growth]_{k, t}. These characteristics are likely to affect the economic performance of prefectures and the election of dynastic legislators. The vector w_{k, t} includes the rate of unemployment, fiscal strength index, total size of population, proportion of population under 15 years old and over 65 years old, proportion of population in the agricultural sector and service sector, degree of urbanization, and population density.Footnote ³² ρ_k controls the effects of unobservable time-invariant characteristics of prefecture k. The summary statistics are reported in Table 6.

Table 6.

Summary statistics for prefecture-level data

Table 7 reports the estimated results. We first check that the amount of fiscal transfers is positively correlated with the GDP growth rate. We include all non-political variables in the equation and estimate the impact of fiscal transfers. Column (1) of Table 7 indicates that the amount of transfers at t − 2 has a positive relationship with the GDP growth rate, while the amount at t − 1 has no statistically significant impact on the growth rate. This result offers evidence in support for our assumption that fiscal transfers have a lagged positive impact on local economic performance.

Table 7.

The effect of dynastic legislators on economic growth at the prefecture level

Notes: Table entries are linear regression coefficients with robust standard errors in parentheses. Standard errors are clustered by prefectures. Estimates are based on the prefecture-level data from 1997 to 2007. The dependent variable is the log rate of total GDP growth. The prefecture and year fixed effects are included in the models. The number of observations is 517. ^∗∗ p < 0.05 and ^∗ p < 0.10.

Next, we turn to our analysis of how the presence of dynastic legislators affects the GDP growth rate. Column (2) reports the estimated result for the third hypothesis. The proportion of rich dynastic legislators with more family assets at t − 3 has a negative and statistically significant impact on the rate of GDP growth. On the other hand, the proportion of poor dynastic legislators with less family assets at t − 3 has a negative but not statistically significant impact. The proportions of rich and poor dynastic legislators at t − 2 have no statistically significant impacts on economic growth.Footnote ³³ Column (3) indicates that the result for rich dynastic legislators holds even after the impact of fiscal transfers is controlled.Footnote ³⁴

The negative coefficient associated with [P.RichDynasty]_{k, t − 3} suggests that dynastic legislators elected from the dynastic district suppress the rate of economic growth by spending the distributive benefits in an inefficient manner. In short, dynastic legislators suppress local economic performance despite the fact they deliver a larger amount of distributive benefits to their districts. More distributive benefits improve local economic performance, yet dynastic legislators spend the distributive benefits inefficiently, which in turn suppresses the growth rate in their districts.