Good friends in high places: Politico-economic determinants of the expropriation and taxation of multinational firms

Abstract

Scholars explaining the conditions that lead governments to expropriate local operations of foreign multinational firms largely focus on how large sunk costs decrease the multinationals’ bargaining power vis-à-vis the host government and how some political regimes (dictatorships) are more inclined to expropriate than others (democracies). Those explanations miss important considerations related to the host-country technological and political environment. In response, we develop and analyze a game theoretical model suggesting that expropriation of multinational firm operations is more likely when: (1) the host-country government capability to monitor taxation of multinational firms is lower; (2) the host-country government capability to run said operations is higher; (3) the host-country government is relatively independent from the exports of the multinational firm-led exports, and (4) political competition is highly restricted. Perhaps paradoxically, we also find that multinational firms are more likely to “self-tax” when host-country governments are too lenient. We illustrate these model-based findings with matched case studies of host-country government interactions with multinational firms in the Venezuelan and Norwegian oil industries of the 20th century.

Appendix: Proof of main results

A.1 Construction of the Function L

Before proving the theorem’s statements, we need to lay the groundwork by characterizing the relevant payoff functions, which in turn requires characterizing the probability of victory. For any two strategies chosen by $\tau _I$ and $s_C$, the levels of support $W_I$ and $W_C$ are defined as:

$$\begin{aligned} W_I= & {} \# \{i: \epsilon _i \ge v_i(\alpha ,s_C) - v_i(\alpha ,\tau _I) - \gamma \}, \\ W_C= & {} \# \{i: \epsilon _i < v_i(\alpha ,s_C) - v_i(\alpha ,\tau _I) - \gamma \}. \end{aligned}$$

Partition the set of citizens into two subgroups according to their economic sector. So, $N_X$ and $N_D$ denote the amount of citizens in sectors X and D respectively ($N_X + N_D = N$). Also, $N_D = \phi N$ and $N_X = (1-\phi )N$. (We shall assume that although $\phi$ is a continuous parameter, $\phi N$ and $(1-\phi )N$ are always integers, a rather trivial assumption that simplifies the analysis significantly.) Exploiting the similarities across citizens, we can then rewrite $W_I$ and $W_C$ as:

$$\begin{aligned} W_I= & {} \# \{i \in D: \epsilon _i \ge v_i(\alpha ,s_C) - v_i(\alpha ,\tau _I) - \gamma \} + \\&\# \{j \in X: \epsilon _j \ge v_j(\alpha ,s_C) - v_j(\alpha ,\tau _I) - \gamma \}, \\ W_C= & {} \# \{i \in D: \epsilon _i< v_i(\alpha ,s_C) - v_i(\alpha ,\tau _I) - \gamma \} + \\&\# \{j \in X: \epsilon _j < v_j(\alpha ,s_C) - v_j(\alpha ,\tau _I) - \gamma \}. \end{aligned}$$

If we define $\epsilon = (\epsilon _1, \ldots , \epsilon _N)$, then the distributions of $W_I$ and $W_C$ are completely defined in terms of $\epsilon$ and the parameters of the utility functions. Then, the probability of victory of the incumbent L is defined simply as the weighted average of all the probabilities of prevailing in the leadership contest, with the weights given by the sizes of $W_I$ and $W_C$. Formally:

$$\begin{aligned} L(\tau _I,s_C) = \int _{W_I} \int _{W_C} \frac{e^{\lambda (W_I - W_C)}}{1+e^{\lambda (W_I - W_C)}} {\rm{d}}W_I(\epsilon \mid \gamma , \phi , \alpha ) \, {\rm{d}}W_C(\epsilon , \mid \gamma , \phi , \alpha ). \end{aligned}$$

A.2. Proof of Theorem 1

To prove this theorem, we carry out the analysis of the game following backward induction, that is, we progress in reverse chronological order along the game’s timeline, each time using the optimal strategy of the latter stages as input in the calculation of the strategies of the earlier ones. So, we consider first the choice of $s_C^*$ for any given $\alpha$. We begin by establishing that there exists a critical value ${\hat{\alpha }}$ such that if $\alpha > {\hat{\alpha }}$, then $s_C^* \ne E$. First, if $\alpha = 1$, $s_C^* \ne E$. This is true because, if $\alpha = 1$, then $L(\tau _I,E) > L(\tau _I,1)$. In fact, if $s_C = 1$, then for any $i \in D$, $v_i(\alpha ,1) \ge v_i(\alpha ,E)$ and at the same time, for any $j \in X$, $v_j(\alpha ,1) > v_j(\alpha ,E)$. Thus, for any realization of vector $\epsilon$, $W_I(\epsilon \mid 1) > W_I(\epsilon \mid E)$. This establishes the result because P(I) is decreasing in $W_I$, a property inherited by $L(\alpha , s_C^*)$. Now suppose that there exists some value $\alpha ' < 1$ such that $s_C^* = E$. The utility of agents $i \in D$ is $y + u(\beta \Pi )$ and for $j \in X$ it is $u(\beta \Pi )$. There exists some value $\alpha ' < \alpha '' \le 1$ such that $\alpha '' \ge \beta$ so that, for that value, $L(\alpha '',1) < L(\alpha '',E)$. This would be true because $v_i(\alpha '',1) \ge v_i(\alpha '',E)$ for $i \in D$ but, at the same time, $v_j(\alpha '',1) > v_j(\alpha '',E)$. By the same token, this proves that the value $\alpha ''$ is increasing in $\beta$ which establishes the theorem’s claim that the function g is increasing in $\beta$. To complete the proof we need to establish two statements. First, we need to prove that $\alpha ^*$ as chosen by I is decreasing in $\gamma$ and increasing in $\phi$. Second, we need to prove that, for any set of parameter values, there is a critical value of $\gamma$ above which $\alpha ^* = 0$. The fact that $\alpha ^*$ is decreasing in $\gamma$ can be proven by studying I’s first-order condition. In particular, the maximization problem

$$\begin{aligned} \alpha ^* = \arg \max _{\alpha } (B - \alpha )L(\alpha ,\gamma ) \end{aligned}$$

results in the condition:

$$\begin{aligned} (B - \alpha ^*) L_{\alpha }(\alpha ^*,\gamma ) - L(\alpha ^*,\gamma ) = 0, \end{aligned}$$

which, upon total differentiation, becomes:

$$\begin{aligned} \frac{{\rm{d}} \alpha ^*}{{\rm{d}} \gamma }= & {} - \frac{[(B - \alpha ^*) L_{\alpha , \gamma } (\alpha ^*,\gamma ) - L_{\gamma }(\alpha ^*, \gamma )]}{[(B - \alpha ^*) L_{\alpha ,\alpha }(\alpha ^*,\gamma ) -(\alpha ^* + 1) L_{\alpha }(\alpha ^*,\gamma )]}. \end{aligned}$$

Since $\gamma$ enters linearly in the functions $v_h(\alpha ,\tau _I) - v_h(\alpha ,s_C)$, then it drops from any higher-order derivative so that $L_{\alpha ,\gamma } = 0$. Instead, first-order differentiation of those functions shows that $L_{\gamma } > 0$. So, the sign of the derivative ${\rm{d}}\alpha ^*/{\rm{d}} \gamma$ is determined by the sign of the denominator. But the denominator is also the second-order condition of the maximization program. This means that, if it is positive, then the program does not admit any interior solution so that $\alpha ^*=0$, which implies ${\rm{d}} \alpha ^*/{\rm{d}} \gamma = 0$. If, on the other hand, the denominator is negative, then $\alpha ^*$ is an interior solution and ${\rm{d}} \alpha ^*/{\rm{d}} \gamma < 0$. This proves the two statements. As regards the impact of $\phi$ on the $\alpha ^*$, the derivative $d(v_h(\alpha ,\tau _I) - v_h(\alpha ,s_C))/d\alpha$ is greater for $h \in D$ than for $h \in X$. Thus, the larger the share of agents in D, the larger the derivative of $L_{\alpha }$, which in turn implies that the larger the value of the optimal $\alpha ^*$.

Proof of Lemma 1

The value $\alpha ^F$ is defined as $\arg \max _{\alpha } v_F((\alpha ,\tau _I^*),s_C^*)$. The term $K_0$ only affects the expected payoff $v_F$ in the event $s_C^* = E$, when $v_F = - K_0$. Denote the probability of said event P(E). Then,

$$\begin{aligned} \frac{\partial ^2 v_F}{\partial K_0 \partial \alpha } = - \frac{{\rm{d}} P(E)}{{\rm{d}} \alpha }. \end{aligned}$$

From Theorem 1 we know that ${\rm{d}} P(E)/ {\rm{d}} \alpha < 0$ so the left-hand side term in this equation is positive. There are two possibilities for $v_F$: it is either quasi-concave, in which case $\alpha ^F$ is an interior solution, or it is not, in which case $\alpha ^F$ is an extreme solution. In the latter case, then the value $\alpha ^F$ is fixed with respect to $K_0$, as is $\alpha _I^*$ (because $K_0$ does not affect the incumbent’s payoff). If, instead, $v_F$ is quasi-concave, then the maximum value is attained in a neighborhood $[{\underline{\alpha }},{\bar{\alpha }}]$ where $d v^F/{\rm{d}} \alpha$ is negative. Then, the value $\alpha ^F$ is increasing in $K_0$.