Before and during Privatization

How different were firms before privatization? To answer this question, we use the reports before the privatization year of each firm. To gain statistical accuracy about firms’ fundamentals, we take three-year averages of the logarithm of assets, logarithm of sales, return over equity, and leverage. We chose these variables because they were available for all firms. In addition, we constructed the changes in these variables by taking the difference between year one and three before privatization to study pre-trends, we constructed Tobin’s q, and we collected the dates when firms were established. We compare these ten variables and the privatization year.

Table 2 compares firms. In addition to firms in our data, we include two other groups: firms without privatization but with reports and firms with privatization but without reports. For the former group, we present summary statistics before the average privatization year in the firm’s industry, but the patterns are similar if we use nearby years. For the latter group, there is unfortunately little information and, therefore, we can only observe their privatization year and industry. The Online Appendix presents the distribution of firms by industry in our data and for all privatizations, where we see that our data over-represent the manufacturing industry and under-represent the wholesale and retail trade industry, but other industries (e.g., electricity and mining) are well represented.

Table 2

FIRMS BEFORE PRIVATIZATION

Notes: Are there observable differences between firms with controversial and uncontroversial privatization processes before privatization? This table provides evidence by presenting averages of variables in the reports before the year each firm was privatized. Column (3) presents the p-value for differences in means across groups in Columns (1) and (2). Column (4) compares distributions in Columns (1) and (2) and presents the p-value from the K-S test (Kolmogorov Reference Kolmogorov1933; Smirnov Reference Smirnov1933). For reference, Column (5) presents descriptive statistics for firms that were not privatized and have annual reports; we use the average privatization year in the firm’s industry. Column (6) presents the privatization year for firms without reports. We present standard deviations in parenthesis and p-values with and without correction for inference in a small sample. More details are given in the Data Construction and Results sections.

Source: Own construction based on firm reports collected by the Superintendencia de Valores y Seguros.

^a Sub-sample of 41 firms.

Each row in Table 2 presents the average and standard deviation of 1 of 11 variables. Columns (1) and (2) examine controversial and uncontroversial privatizations separately. Column (3) presents p-values for differences in means across groups, with and without correction for small sample inference.^{Footnote 14} Column (4) uses the Kolmogorov–Smirnov (K-S) test to compare the distribution of variables across groups (Kolmogorov Reference Kolmogorov1933; Smirnov Reference Smirnov1933). Columns (1)–(4) show few statistically significant differences in profitability, growth prospects as measured by Tobin’s q and asset growth, indebtedness, or firm age before privatization. The exception is firm size; we observe controversial firms were, on average, smaller. Although our ability to detect differences across firms may be affected by the sample size, the majority of differences are also of relatively small economic magnitude.^{Footnote 15} When compared to firms in our data, Column (5) reveals that firms privatized by the regime were significantly larger, older, and had lower performance, but had similar debt compared to other firms with reports but not privatized.

To improve our understanding of privatization characteristics, Table 3 presents different regression specifications using underpricing (Columns (1)–(4)) and closeness to the regime (Columns (5)–(7)) as separate dependent variables, and pre-privatization variables and industry fixed effects as predictors. All in all, we observe that firms sold in the 1980s exhibited significantly more underpricing (p-value < 0.05) and the underpricing was larger when the buyers were close to the regime (p-value < 0.10). In contrast, the importance of pre-privatization variables is economically smaller. These results suggest that there might be a relationship between both privatization characteristics that we discuss below.

Table 3

UNDERSTANDING PRIVATIZATION CHARACTERISTICS

Notes: What pre-privatization variables predict privatization characteristics? This table presents estimates from cross-sectional regressions using privatization characteristics as dependent variables, that is, underpricing or closeness to the regime, and pre-privatization variables as predictors. More details are given in the Data Construction section. Robust standard errors are in parentheses. Significance levels: *** p < 0.01, ** p < 0.05, * p < 0.1.

Source: Own construction based on firm reports collected by the Superintendencia de Valores y Seguros.

In sum, we interpret Tables 2 and 3 as evidence that, although the privatization decision may have been driven by firm dynamics, the type of privatization—that is, controversial versus uncontroversial—seems not to have been driven by firms’ behavior. In what follows, we present several econometric exercises supporting this interpretation.

The Stock Market

We use the framework of Fisman (Reference Fisman2001) to test whether firms with controversial processes benefited financially from Pinochet. We study the stock market value of controversial firms after an exogenous shock that increased the probability of political transition.^{Footnote 16} If controversial firms benefited from the dictatorship, we expect to see a decrease in their value after the unexpected outcome of the referendum that ended the dictatorship. This referendum was held on October 5 of 1988 and had Pinochet running to remain in office for the next eight years (with yes or no votes). The regime wanted to validate themselves as democratic in front of the international community. Both the rejection of Pinochet’s continuation and the regime’s acknowledgement of results were unexpected.^{Footnote 17}

To measure changes in the stock market, we digitize daily stock prices of listed firms from the newspaper El Mercurio, available at Chile’s National Library. We restrict attention to firms that were traded for at least four months before the plebiscite to analyze abnormal returns (i.e., the difference between returns and expected returns):

(1) $$A{R_{it}} = {R_{it}}-\left( {{a_i} + {b_i}{R_{mt}}} \right),$$

where R _it is the stock return of firm i on day t, R _mt is the market return on day t, and we estimate the parameters a _i and b _i using pre-plebiscite data. As for robustness, we also looked at cumulative abnormal returns, defined as Σ_t=0^t=jAR _it (see Campbell et al. Reference Campbell, Lo and Craig MacKinlay1997 for details). The usage of pre-plebiscite data leaves us with 41 firms, 20 of which were controversial. We present estimates of the following regression graphically:

(2) $$CA{R_{ijt}} = {b_t}Controversia{l_i} + {d_t}{X_i} + {n_{jt}} + {e_{ijt}},$$

where CAR _ijt = Σ_k = 0^tAR _ik is the cumulative abnormal return for firm i, which operates in industry j, from the day of the plebiscite up to t days; Controversial _i is an indicator for controversial firms; X _i represent pre-privatization controls (i.e., assets, sales, return over equity, and leverage); n _jt is a set of industry fixed effects; and e _ijt is a mean zero error term. The parameter of interest is b _t and measures the differential cumulative abnormal return for firms with controversial privatizations. All parameters in Equation (2) are indexed by t because we estimate it separately for t = 1, 3, 5, 8, and 10.

Figure 3(a) presents daily abnormal returns by type of privatization, and Table 4 the corresponding regression estimates. Consistent with our hypothesis, we find a statistically significant decrease in abnormal returns among controversial firms the day after the plebiscite. The drop in abnormal returns corresponds to approximately 7.5 percentage points (Column (1) of Panel A in Table 4, p-value < 0.01), it lasts for at least ten days, and it is robust to the inclusion of pre-privatization controls. Moreover, these results are similar when accounting for a potential non-normality in the distribution of abnormal returns and the cross-sectional correlation between stocks in the day of the event (Kolari and Pynnönen Reference Kolari and Pynnönen2010). When compared to prominent estimates in the literature, we calculate that this drop in returns is larger than the one in Fisman (Reference Fisman2001) and similar to those in Ferguson and Voth (Reference Ferguson and Voth2008) and Acemoglu et al. (Reference Acemoglu, Johnson, Kermani, Kwak and Mitton2016).

Figure 3.

THE STOCK MARKET

Notes: Own construction using stock price data hand collected from the contemporary newspaper El Mercurio, available at Chile’s National Library. Details are given in the Results section.Sources: Own construction based on firm reports collected by the Superintendencia de Valores y Seguros and stock prices published by the newspaper El Mercurio.

Table 4

THE STOCK MARKET

Notes: Does the value of firms with controversial privatization processes change after the unexpected announcement of Chile’s transition to democracy on October 5, 1988? Each column in Table 4 provides evidence by presenting ordinary least squares (OLS) estimates (Equation (2) in the paper). We collected data on stock prices from the newspaper El Mercurio. Our sample decreases from 50 to 41 firms, because in order to calculate abnormal returns,we need to observe stock prices four months before the event we study and we do not observe these for 9 firms. More details are given in the Results section. Robust standard errors are in parentheses, and p-values correcting for small sample inference are in square brackets. Significance levels: *** p < 0.01, ** p < 0.05, * p < 0.1.

Source: Firm reports collected by the Superintendencia de Valores y Seguros and stock prices published in the newspaper El Mercurio.

Figures 3(b)–3(e) show that these patterns are particular to the announcement of the transition. We observe similar abnormal returns around other important political events, namely, the day when Pinochet was nominated to be on the ballot at the plebiscite (30 August 1988), the last constitutional reform in dictatorship (30 July 1989), the 1989 presidential election (14 December 1989), and when the new government took office (3 March 1990). As highlighted in prior literature, the behavior of Chilean investors is also consistent with controversial firms having disproportionately benefited from their connections to Pinochet’s regime.

The Credit Market under Dictatorship

The credit market can reveal whether firms with and without controversial privatizations were receiving a differential treatment from the regime. To study this market, we use information about firms’ outstanding debt with Banco del Estado (Bank of the State, the only state-owned bank in the country) and other types of banks.^{Footnote 18} We study firm debt financing with these banks in the period between October 1988 and March 1990. In particular, we estimate the following regression before and after the plebiscite:

(3) $${Y_{ijt}}^k = {b_t}^kControversia{l_{ij}} + {d_t}^k{X_{ij}} + {n_{jt}}^k + {e_{ijt}}^k$$

where i indexes firms, j industries, t periods, and k the type of bank (i.e., state-owned, private, or international). The dependent variable Y _ijt^k is an indicator for firms with outstanding debt with bank k in period t, the average interest rate with this bank, leverage, or the logarithm of total debt. We study two periods, before and after the plebiscite: 1986–1987 and 1988–1990. All regressions include pre-privatization controls X _ij (i.e., assets, sales, return over equity, and leverage) and industry fixed effects by period, n _jt. The coefficients of interest b _t^k measure the within-industry differences among controversial privatizations in the outcome of interest while controlling for pre-privatization differences. Note that we allow coefficients to differ by period and bank type.

Table 5, Panel A presents estimates of Equation (3) after the plebiscite. Column (1) shows that controversial privatizations were 30 percentage points more likely to have loans from Banco del Estado between 1988 and 1990 (p-value < 0.05), when it was known Pinochet would be leaving. This result is consistent with the findings in Khwaja and Mian (Reference Khwaja and Mian2005) and suggests that the dictatorship used the credit market to benefit these firms; it is also consistent with the evidence in González and Prem (Reference González and Prem2019), which finds that firms in Pinochet’s social network obtained more loans from state-owned banks between 1988 and 1990. Column (4) shows that these loans had 4 percentage points lower interest rates, but this estimate relies on a smaller sample and it is only statistically significant when using small sample inference. In contrast, we do not observe any of these patterns between controversial firms and private or international banks and point estimates are of significantly smaller economic relevance (Columns (2), (3), (5), and (6)). Finally, Column (7) shows that there are no statistically significant differences in leverage between privatizations, which suggests firms either substituted loans across banks or increased their equity in this period, and Column (8) shows that controversial firms had more total debt.

Table 5

THE CREDIT MARKET

Notes: Each column presents estimates of Equation (3) where we measure the dependent variable in 1988–1990 (Panel A) or in 1986–1987 (Panel B). Dependent variables measuring loans, interest rates, and leverage (debt over assets) are own construction from firm-level reports. Banco del Estado is the main state-owned bank in Chile. More details are given in the Results section. Robust standard errors are in parentheses, and p-values correcting for small sample inference are in square brackets. Significance levels: *** p < 0.01, ** p < 0.05, * p < 0.1.

Source: Own construction based on firm reports collected by the Superintendencia de Valores y Seguros.

Although the reader might be concerned that controversial privatizations were potentially different in unobservable dimensions, and this is the reason why we observe a different credit market for these firms, the evidence suggests this was probably not the case. Table 5, Panel B presents estimates of Equation (3) using reports before the plebiscite and we do not find statistically significant differences in state loans or interest rates. Moreover, most point estimates are economically smaller than in Panel A and patterns with other banks are again similar across types of firms. Interactions between controversial firms and the state bank seem to have changed over time, but we can only reject the similarity of coefficients in Column (1) of Panels A and B with a p-value of 0.11 when we use a pooled panel specification. Point estimates also suggest total debt increased in 1988–1990, but we cannot reject that it remained similar. Because of this and other concerns below, we discuss additional robustness checks.

The Beginning of Democracy

Controversial privatizations differed significantly at the very beginning of democracy. We consider a version of Equation (3) with time-invariant coefficients and measuring the dependent variable in 1990. To be consistent, we consider the same four firm-level outcomes (assets, sales, return over equity, and leverage) and also stock returns since the year of privatization. Note that we again control for pre-privatization variables and include industry fixed effects in our estimation.

Table 6 presents the results. Columns (1) and (2) show that controversial firms grew faster than other firms in the same industry during the dictatorship. Given that we are controlling for previous size and these firms were smaller, this result means that controversial firms partially caught up in terms of size. Results using the logarithm of sales as a dependent variable confirm this faster growth, although the point estimate is not statistically significant. In contrast, Columns (3) and (4) show that there was little difference in indebtedness levels (i.e., leverage) and profitability (i.e., return over equity). Finally, Column (5) shows that stock returns since the year of privatization were statistically similar between controversial and uncontroversial firms, although the point estimate suggests that the stock returns of controversial firms were lower.

Table 6

THE BEGINNING OF DEMOCRACY

Notes: Each column in Table 6 presents estimates of Equation (3) with outcomes at the beginning of democracy, that is, at the end of the year 1990. More details are given in the Results section. Robust standard errors are in parentheses, and p-values correcting for small sample inference are in square brackets. Significance levels: *** p < 0.01, ** p < 0.05, * p < 0.1.

Source: Own construction based on firm reports collected by the Superintendencia de Valores y Seguros and stock prices published in the newspaper El Mercurio.

Overall, the results reveal that firms with controversial privatizations grew significantly more in dictatorship but experienced little improvement in their profitability.

Politics in Democracy

Are controversial firms influencing politics in democracy? We focus on three sources of distortions: the employment of politicians, the financing of political campaigns, and tax avoidance. We begin by studying employment of politicians as board members, empirically associated with rent extraction (e.g., Khwaja and Mian Reference Khwaja and Mian2005; Goldman, Rocholl, and So Reference Goldman, Rocholl and So2013) and thus an important source of misallocation (e.g., Cingano and Pinotti Reference Cingano and Pinotti2013). Because misallocation affects total factor productivity (Hsieh and Klenow Reference Hsieh and Peter2009), understanding the formation of these connections is critical.

We study the evolution of political connections in a dynamic fashion. We estimate Equation (3) using as the dependent variable an indicator for firms that employed at least one politician for their board. To capture dynamics, we measure this employment in different points in time and use three types of politicians: (1) former politicians of the Pinochet regime who enjoyed significant political power at the beginning of democracy and who we call “politicians of the old regime”; (2) politicians of the new democratic incumbent coalition opposed to Pinochet called Concertación, who we call “politicians of the new regime”; and (3) any of the previous politicians, who we call “any politician.”

Table 7 shows that controversial firms were 25 percentage points more likely to employ any politician after the dictatorship, 25 percentage points more likely to employ a politician from the Pinochet regime at the beginning of democracy, and 40 percentage points more likely to employ politicians of the new regime after 15 years of democracy. These coefficients represent economically large magnitudes and the dynamics are revealing. Controversial firms substituted connections from the old to the new regime after a decade in democracy. These connections reverted almost perfectly, and in 2005, we observe more than half of the controversial firms in our data having connections to the new democratic coalition. In contrast, politicians of the old regime were no longer in these firms by 2005.

Table 7

POLITICS IN DEMOCRACY

Notes: Each row in Table 7 presents estimates of Equation (3). The “old regime” corresponds to the Pinochet regime (1973–1990), and the “new regime” corresponds to the period after 1990. More details are given in the Results section. Robust standard errors are in parentheses, and p-values correcting for small sample inference are in square brackets. Significance levels: *** p < 0.01, ** p < 0.05, * p < 0.1.

Sources: Own construction based on firm reports collected by the Superintendencia de Valores y Seguros and campaign finance data published by Servicios de Impuestos Internos.

Controversial firms may also distort the political arena via financing political campaigns. This is the case studied in Claessens, Feijen, and Laeven (Reference Claessens, Feijen and Laeven2008), which shows that Brazilian firms that contributed to political campaigns had higher stock returns because they benefited from preferential access to bank financing. In addition, political contributions have also been found to affect public procurement (Baltrunaite forthcoming). Although perhaps intuitive, this type of analysis has been relatively scarce because data on campaign contributions can be difficult to obtain.

The list of firms that illegally financed political campaigns was revealed after an extensive investigation by the Chilean tax authority. Accusations of illegal campaign financing before the presidential election of 2013 were the motivation behind that investigation. The illegality of these transfers took the form of monetary payments from firms to politicians for “services” that were never delivered. These interactions were summarized, and the list of firms participating was publicized in the press. We also observe the list of firms that contributed to campaigns legally between 2005 and 2013. We construct two indicator variables, one for illegal and another one for legal campaign finance. We observe that 46 percent of firms in our data legally contributed to political campaigns in the period between 2005 and 2013, and 22 percent contributed illegally in 2013.

We follow the same econometric strategy and estimate Equation (3) using an indicator for legal or illegal campaign finance as the dependent variable. The last rows in Table 7 present results. Estimated coefficients show that controversial privatizations were 31 percentage points more likely to legally finance political campaigns (p-value < 0.05) and 19 percentage points more likely to contribute illegally, although the latter result is not statistically significant at conventional levels (p-value 0.19). These differences are economically meaningful. On one hand, only 37 and 19 percent of uncontroversial privatizations contributed legally and illegally (Column (3)). On the other hand, more than 68 and 37 percent of controversial privatizations did. These results suggest that controversial firms indeed seem to have attempted to exert influence in the political arena.

The last row in Table 7 shows that firms with controversial privatizations employed board members in democracy who were 36 percentage points more likely to appear in the Panama Papers (p-value 0.02). This difference is large, as more than half of controversial firms employed at least one board member who appeared in these documents. In contrast, only 18 percent of uncontroversial firms employed a board member from the list. We highlight that this is a legal behavior, but it nevertheless decreases tax revenues and it is therefore important to study.

Robustness and Omitted Variables

A variety of econometric exercises suggest our findings are robust and the effect of unobservables is minimal. We begin by showing similar estimates when we include additional control variables or exclude particular firms from the estimation. Additionally, the effects of controversies are similar or, if anything, larger if we use the processes studied by Marcel (Reference Marcel1989) and Mönckeberg (Reference Mönckeberg2001) to define controversial privatizations. Finally, we show results are also robust to the use of modern matching estimators and techniques that adjust for the effect of unobservables, suggesting omitted variables are not driving our results. Table 8 presents all of these additional results.

Table 8

ROBUSTNESS OF RESULTS AND OMITTED VARIABLES

Notes: Each estimate comes from a different estimation strategy. See the Results section for details. Robust standard errors are in parentheses. Significance levels: *** p < 0.01, ** p < 0.05, * p < 0.1.

Sources: Own construction based on firm reports collected by the Superintendencia de Valores y Seguros and stock prices published by the newspaper El Mercurio.

We begin showing robustness to additional controls. Scholars have argued the two waves of privatization were different from each other, as the former privatized firms nationalized by Salvador Allende and the latter privatized long-standing state-owned firms. To check for this potential confounding factor, we constructed an indicator that identifies the “privatization wave” of a firm and included it as an additional control. Column (4) shows that the results controlling by wave are similar. Another potential confounder could be a change in the controller of a firm, but Column (5) shows similar results if we eliminate the few firms that changed controllers between 1990 and 2005.^{Footnote 19}

A different classification of firms and the exclusion of single firms from estimation provide more robustness to the results. First, our clustering algorithm could have captured unobservable variables, so it is important to check if results are driven by the procedure we chose. Besides using other clustering algorithms, we also classified firms as controversial if these were mentioned as corrupt by Marcel (Reference Marcel1989) or Mönckeberg (Reference Mönckeberg2001), who argue 8 of our 50 firms were sold underpriced.^{Footnote 20} Column (7) in Table 8 shows results are larger using their classification. Second, we checked if results changed when we exclude one firm at the time from the estimation. Results in the Online Appendix confirm that our estimates are not driven by single observations.

Another threat is the omission of variables that could be correlated with controversies and explain the outcomes of interest. Two econometric techniques suggest the estimates are robust and the effect of omitted variables is minimal. First, we use matching techniques to perform improved comparisons. Operationally, we calculate the probability of controversies in a privatization using pre-privatization variables and industry fixed effects. Then, we perform three estimations, one in which we follow Crump et al. (Reference Crump, Joseph Hotz, Imbens and Oscar2009) and restrict the sample to firms that have similar probabilities of controversies (Table 8, Column (1)), another in which we simply control for the probability of controversies (Column (2)), and a last one in which we create a counterfactual for each firm using the k-nearest neighbors (Column (3)).^{Footnote 21} The second strategy uses the predictive power of observable variables to adjust the coefficient of interest by considering the effect of unobservables. This “coefficient stability approach”—first proposed by Altonji, Elder, and Taber (Reference Altonji, Elder and Christopher2005) and refined by Oster (Reference Oster2019)—again delivers similar estimates (Table 8, Column (6)).

We conclude that the evidence suggests the existence of a preferential treatment flowing from the regime to controversial privatizations, and these firms operated as political corporations in the democracy period.