M&A transactions allow firms to consolidate their long-term strategic objectives. The business combination can provide market power in the same industry or in the formation of conglomerates, gains from synergy benefits in the form of greater growth, or economies of scale, and positive effects on returns, placing the target firm under a more experienced management (Damodaran, 2008). In this way, M&A activities can move the economy and generate interest for analysts, academics, and policymakers (Bonaime et al., 2018).

Wood et al. (2004) listed some reasons for the practice of M&A, which can be organized into two groups: (i) Strategic reasons, involving anticipation of a move by competitors; intensity of competition, with the emergence of new entrants and substitutes; and the need to generate economies of scale. (ii) Political and institutional reasons, which involve the influence of shareholders and other primary stakeholders such as government and business partners; political motives within the organization; and the tendency of companies to follow one another, leading to mimetic behavior.

This last behavior is particularly interesting in the context of this research, as taking other organizations as a model can constitute a response to uncertainty (DiMaggio & Powell, 1983). Some studies predicted a positive association between uncertainty and M&A transactions (Duchin & Schmidt, 2013; Sha et al., 2020).

On the other hand, the most conventional approach used to explain this relationship is the Real Options Theory (Bernanke, 1983; Dixit & Pindyck, 1994), which predicts a negative relationship. This theoretical strand signals that real corporate investments react negatively to uncertainty, because, due to its irreversibility, firms may prefer to keep their cash holdings for the purpose of prevention and/or speculation and risk reduction (Duong et al., 2020). In these circumstances, firms would have incentives to postpone their acquisitions, as the option of waiting for new information is valued in this context (Bonaime et al., 2018). New studies contribute to this current discussion, relating a specific source of uncertainty, in its political dimension, due to the EPU metric developed by Baker et al. (2016).

Such empirical studies have confirmed the theoretical hypothesis that economic policy uncertainty can often delay corporate investment (Akron et al., 2020; Chen et al., 2020; Gulen & Ion, 2015; Wang et al., 2014), with some exceptions (Liu et al., 2020). Specifically with Mergers and Acquisitions, using a sample of American companies over the period from 1986 to 2014, Nguyen and Phan (2017) identified that economic policy uncertainty is negatively related to the propensity to acquire other firms and is positively related to the time taken to complete the deals. The authors also found that uncertainty motivates acquiring firms to use shares as a form of payment and to pay lower acquisition premiums.

Moving along this line, Bonaime et al. (2018) found a strong negative association between economic policy uncertainty and M&A activity with a specific sample of US companies and, at a macroeconomic level, considering all the announcements made in the country. Consistent with the Real Options Theory, the authors identified that this effect is intensified for less reversible trades. On the other hand, the effect is mitigated for businesses that cannot be delayed due to the level of competition. The analysis at the macroeconomic level, based on a Vector Autoregressive model (VAR), showed that both the total value and the total number of transactions respond negatively to a shock of economic policy uncertainty, with a persistent effect for up to 12 months ahead.

Replicating the study by Bonaime et al. (2018), Borthwick et al. (2020) investigated this relationship in the Chinese M&A market. The authors identified that economic policy uncertainty reduced the probability of mergers and acquisitions in the following year among Chinese companies, confirming the existence of a negative effect.

Thus, consistent with the stream of empirical evidence and discussions on the general relationship between the variables of interest in this study, the following research hypothesis is proposed:

(H1) Publicly traded Brazilian companies reduce their merger and acquisition activity in response to heightened economic policy uncertainty.

Uncertainty has been the target of several contributions in macro finance (Barboza & Zilberman, 2018; Godeiro & Lima, 2020; Pereira, 2001; Souza et al., 2019; Zerbinatti et al., 2021). In the Brazilian scenario, Barboza and Zilberman (2018) investigated the impacts of uncertainty on economic activity and attested that there is a contractionary effect of both domestic uncertainty (in greater intensity) and external uncertainty. In that same scenario, the countercyclical pattern of uncertainty was evidenced, and it was shown that increases in its magnitude precede economic crises (Godeiro & Lima, 2017). Using conditional volatility models to build uncertainty proxies from macroeconomic variables, studies from the early 2000s already showed that investment is negatively affected by uncertainty, corroborating the economic theory for Brazil (Pereira, 2001).

With the evolution of computational capacity recently, associated with criticisms about the use of uncertainty proxies based on volatility (Godeiro & Lima, 2017), proper measures of uncertainty could be developed from techniques that allow capturing it in more specific dimensions, such as economic policy (Baker et al., 2016), which is featured in this manuscript.

The conduct of economic policy by the government impacts the behavior of the financial market and companies, which must respond to government actions. Such an impact can be intensified when there is uncertainty about who will make policy decisions, when and what decisions will be made, and their subsequent effects on the economy (Baker et al., 2016). The inability to predict these characteristics can be signaled with preventive and/or precautionary decisions to cushion the shocks derived from uncertainty and reduce risks. Companies can increase their cash holdings (Demir & Ersan, 2017; Duong et al., 2020), increase their payout levels (Attig et al., 2021), reduce funding levels (Zhang et al., 2015) and delay their investments (Akron et al., 2020; Gulen & Ion, 2015), including M&A transactions (Bonaime et al., 2018; Nguyen & Phan, 2017).

Uncertainty is a construct not directly observable and therefore difficult to measure. It is often captured by the dispersion of macroeconomic expectations and asset price volatility in the financial market. Nonetheless, Baker et al. (2016) proposed an indicator for this quantification, which is based on counting the frequency of news in newspapers that report uncertainty in the political scenario. Originally created for the United States, the EPU has three types of underlying components: (i) the first is a textual analysis component, derived from search results in the ten largest newspapers in the country, to get the average monthly news count that contains the terms “uncertain” or “uncertainty” and “economic” or “economy”, along with other relevant political terms: “congress”, “deficit”, “federal reserve”, “legislation”, “regulation”, or “white house” (including more variants for all terms); (ii) the second component is based on reports from the Congressional Budget Office (CBO) which compiles lists of temporaryfederal tax code provisions, given that temporary fiscal measures are a source of uncertainty for businesses and households; and (iii) the third is based on market analysts' forecasts dispersion on future levels of the consumer price index and government spending at the federal, state, and local level (Baker et al. 2016).

All components are disclosed separately and in aggregate on the website https://www.policyuncertainty.com/. According to Baker et al. (2016), the extension of the measure over time, and, for the other countries, focused only on component (i) of journalistic media. For that reason, they can also be called Newspaper-based EPU. The EPU for Brazil is officially calculated and published on the policyuncertainty portal with the component (i), only, with adaptations to the local reality in the political terms sought, using archives from the newspaper “Folha de São Paulo” since 1991.

Alternatively, Brazil has the Economic Uncertainty Index (IIE-Br), which was developed by Ferreira et al. (2019) and measures, however, the general levels of economic uncertainty. The IIE-Br is produced by the Brazilian Institute of Economics - IBRE/FGV and comprises two components, the (i) media component (80% weighted) with the frequency of articles mentioning economic uncertainty in the six largest high-circulation newspapers in the country, namely: “Valor Econômico”, “Folha de São Paulo”, “Correio Brasiliense”, “Estadão”, “O Globo” and “Zero Hora”. To address economic uncertainty, textual analysis comprises the terms “ECON” for economy and “INSTAB”, “UNERT” and “CRISES” for uncertainty. The second specific component of that metric comprises an indicator of market analysts’ forecasts dispersion on macroeconomic variables: Basic interest rate (Selic), Broad Consumer Price Index (IPCA) and exchange rate (PTAX) (Ferreira et al., 2019).

Both EPU Brazil and IIE-Br capture the volatility of the perception of uncertainty in their political and economic dimensions, respectively. A greater variability of the EPU index is noticed, when comparing the historical evolutions of the two metrics. Despite being measured with a similar methodology, the differences in their magnitudes are consistent with their purposes and calculation methods, in addition to the fact that different periods were used for its standardization (the EPU Brazil starts in 1991, while the IIE-Br starts in 2000). Furthermore, the EPU has a greater emphasis on economic policy and may suffer from some perspective bias from the only news source it includes, which may contribute to its greater volatility (Schymura, 2019). Figure 1 below shows the evolution of the historical series for the two indices:

Figure 1. Source: Data available at policyuncertainty.com.

We noticed that both indexes have a persistent upward trend from 2015 on. This increase can be explained by the deterioration of the political situation in the period, which led to several events involving political actors, in addition to the loss of investment grade, with the downgrade of Brazil’s credit rating by Standard & Poor’s (Schymura, 2019).

Thus, we use the two indicators alternatively as proxies of uncertainty in the following econometric analysis. With this, we expect to see possible differences in the responsiveness of M&A transactions, in relation to uncertainty in the political and economic dimension in Brazil.

To reach the objectives, this investigation proposes models that estimate the propensity for M&A, based on studies by Bonaime et al. (2018), Borthwick et al. (2020), Nguyen and Phan (2017) and Sha et al. (2020), with adaptations for the Brazilian market and for the available data. To this end, the binomial logistic regression was used to estimate the parameters of the model, for predicting the probabilities of an acquisition announcement in the following year (t+n) as a function of EPU and IIE-Br indexes in t, as well as control variables at the firm, industry, and macroeconomic levels, for a sample of publicly traded companies.

Initially, firm-level data was collected from a sample of 172 publicly traded non-financial companies listed on the Brazilian stock exchange (B3). Companies in the financial industry were excluded, following previous studies (Borthwick et al., 2020) in order to maintain the possibility of comparability among them. The number of companies is the result of a previous filter, from the Economatica® database, which considered a total of 313 non-financial companies over 9 years, from 2010 to 2018. Of this total, 141 companies with uncovered liabilities or did not disclose financial information in all fiscal years were excluded, which could hamper the collection of indicators. At this point, we reached 1,548 observations of companies/year. It was also verified that not all these companies have an active presence in the trading sessions of the stock exchange, which could jeopardize the calculation of its market ratios. With this, we delimited the observations to those in which the company had a presence greater than 40% in the year. This caused 44 companies to be removed from the analysis, as they had a presence below this level in all years. Thus, the final number of companies analyzed was 128, comprising 943 observations of companies/year with complete information in an unbalanced panel. The pooled logit model was used in the regression analysis of the independent variables on the defined dependent variable.

The dependent variable in this study (MA_it+1) takes the binary form, where it receives the value one if the firm announces an acquisition in the subsequent period (t+1) and zero otherwise. The companies' mergers and acquisitions data were extracted from the SDC Platinum® database (Refinitiv®) until the year 2019, the date on which the database was available to the authors at the time of collection. The independent variable is the economic policy uncertainty, represented by the EPU metric by Baker et al. (2016) and alternatively the IIE-Br metric by Ferreira et al. (2019). The uncertainty proxies were both extracted from the Economic Policy Uncertainty Index portal (policyuncertainty.com). The data for the construction of the other macroeconomic control variables were extracted from the Time Series Management System - SGS of the Central Bank of Brazil (bcb.gov.br/sgspub). As there is a time lag of the explanatory variables in relation to the dependent variable, the data series for the former extend to 2018.

The explanatory and control variables at the firm, industry and macroeconomic levels are detailed next (Table 1).

Table 1

Explanatory Variables	Description	Source
LNEPU	Natural logarithm of the weighted average between the months of each year of the EPU Brazil index.	Nguyen and Phan (2017); Bonaime et al. (2018); Borthwick et al. (2020); Sha et al. (2020).
LNIIE	Natural logarithm of the weighted average between the months of each year of the IIE Brazil index.
Firm level
LNSIZE	Natural logarithm of total assets.
ROA	Ratio of earnings before interest and taxes to total assets.
VSALES	Change in net revenue in relation to t-1.
LEV	Ratio between total debt and total assets.
CX	Ratio between cash and equivalents and total assets.
MTB	Market-to-Book Index. Ratio between market value and book value of equity.
RET	Buy-and-hold stock return during period t trading days.
VOL	Annualized standard deviation of daily stock returns during period t.
Industry level
IMTB	Median of the Market-to-Book index for each industry in period t
IRET	Median of returns for each industry in period t
IVOL	Median of annualized standard deviations of daily returns for each sector in period t
HHI	Herfindahl-Hirschman Index: sum of the square of the market shares of companies in the industry.
Macroeconomic level
INVOP	Investment opportunities: first principal component extracted from the linear combination between four indexes: Consumer Confidence Index - ICC; Economic Activity Index - IBC; Future Expectations Index - IEX; and yearly change in GDP .
SELIC	Yearly variation of the Selic rate.

Note: the calculation of industry level variables was not restricted to the companies in the sample. In this case, all listed companies with available data in the industry in period t were considered to calculate the medians and the HHI. In calculating the macroeconomic level variable INVOP, the year average was considered for the ICC, IBC and IEX indexes, that are released monthly by the Central Bank of Brazil (https://www3.bcb.gov.br/sgspub/). Details of the PCA results are provided in Appendix A. Due to limitations in the available data, some variables used to calculate INVOP differ from those used by Bonaime et al. (2018).

For the explanatory variables of uncertainty, which are calculated and reported monthly, the weighted average was applied, considering greater weighting in the last month, as the level of uncertainty of the most recent month can have a greater impact on decisions (Nguyen & Phan, 2017; Schwarz & Dalmácio, 2020). Alternatively, the arithmetic mean was also used, and its significance indicated in an explanatory note to the table.

All firm-level variables were winsorized at the 1st and 99th percentiles, following the guidelines by Bonaime et al. (2018) and Borthwick et al. (2020). Furthermore, considering that industry common factors of companies can affect acquisitions, we have included industry fixed effects controls in some models, classified according to macro industry segmentation of B3. Following Nguyen and Phan (2017), we did not control for year fixed effects, since all firms are subject to the same political uncertainty in a given year and this could absorb the explanatory power of the variable of interest (Gulen & Ion, 2015; Nguyen & Phan, 2017). Models include robust standard error estimates with clustering criteria by year (Bonaime et al., 2018; Borthwick et al., 2020; Nguyen & Phan, 2017).

Both EPU and M&A volume show a cyclical movement over time and can be simultaneously correlated with unobservable factors at the macro level (Nguyen & Phan, 2017). Such a fact can exacerbate concerns about the existence of endogeneity in the model (Hill et al., 2021). One way to deal with this is running the model with instrumental variables (IV). In similar reasoning to previous empirical studies (Nguyen & Phan, 2017; Sha et al., 2020; Schwarz & Dalmácio, 2020), the variables yearly level of “Chamber of Deputies support to the president’s position” (Estadão, 2022) and US EPU were used alternatively as instruments. However, we were not successful in instrumentalizing the uncertainty variables, which is a limiting factor of this study. In fact, it can be a challenge to find valid instruments in management research (Hill et al., 2021) and such consideration remains as a suggestion for further studies.

As we deal with the modeling of a binary variable and, therefore, a limited dependent variable (Maddala, 1986), this analysis should employ a suitable econometric modeling. Among the possibilities of models for this specificity, Wooldridge (2019) highlights the Linear Probability Model (LPM, which uses the Ordinary Least Squares estimator) and the Probit and Logit models (that employ the Maximum Likelihood estimator), which are non-linear models. The empirical literature on M&A uses these three possibilities to estimate the acquisition propensity (Bonaime et al., 2018; Erel et al., 2021; Nguyen & Phan, 2017) depending on some modeling specificities, such as, for instance, the existence of interaction terms and/or many dummy variables among the explanatory variables, in which, in this case, the LPM would be indicated to mitigate the problem of incidental parameters, which can occur with non-linear models (Erel et al., 2021; Nguyen et al., 2020). Probit and Logit models often result in qualitatively similar estimates and for this analysis, we chose to follow the approach adopted by Bonaime et al. (2018) using Logit. The Logit model employed can be specified according to the functional form (Wooldridge, 2019):

PMAit+1=1|x=G(β0+xβ) (1)

where,

Gβ0+xβ=11+e-β0+xβ=11+e-Z=eZ1+eZ (2)

and

xβ=λUncertaintyt+δFirmit+γIndustryst+ωMacrot+dFE_Industrys (3)

Where the model response variable is a probability of becoming an acquirer in the following year (MA_it+1 = 1) conditional on a vector x that denotes the set of explanatory and control variables, described in the previous section, and varies between 0 and 1. G is the cumulative logistic distribution function that takes values strictly between 0 and 1 and ensures that the estimated probability is limited to this range. β₀ is the constant term. β is a vector that denotes the set of estimated parameters for the explanatory and control variables: λ, δ, γ, ω, d. Uncertainty_t represents the interest explanatory variable that assumes EPU and IIE-Br indexes, alternatively. Firm _it , Industry_st , and Macro_t are the set of firm, industry, and macro-level control variables, respectively. FE_Industry_s denotes the use of dummies for industry fixed effects control. The subscripts i,s and t indicate, for its respective vector, that the variables vary between companies i, between industries s and/or between years t.

It is important to note that the interpretation of the coefficients of the Probit and Logit models is not straightforward. A priori, the sign of the coefficient is interpreted, but not its magnitude, due to the non-linear nature of the function G (Wooldridge, 2019). For this purpose, the marginal effect of the variable x must be reported, which measures the change in the probability of success of y (y = 1) given a unit change in x. For this, we need to resort to the partial derivative of the function G, given p(x) = P(y = 1|x) (Wooldridge, 2019):

∂p(x)∂xj=G'β0+xββj=eZ1+eZ 11+eZβj  eZ1+eZ2 βj (4)

Or, simplifying, given the equality in Eq. (1), we have that P=eZ1+eZ and 1-P=11+eZ. Thus, substituting the terms of Eq. (4) the calculation of the marginal effect for the variable j:

βjP(1-P) (5)

In which subscript j refers to the parameter β estimated for the j-th independent variable. In this way, it is possible to interpret the effect of oscillations in the variable x on the probability of success of y.

Table 2 presents the summary statistics for the sample of companies and for the industry and macroeconomic variables in the studied period. The average political and economic uncertainty levels for the period were 171.419 (Ln: 5.114) and 104.766 (Ln: 4.652) respectively. There is greater volatility for the EPU, a behavior also perceived in Figure 1. The average assets (AA) of these companies in the period was BRL 6.66 billion, approximately (Ln: 15.712). The companies’ average annual ROA was 4.5% and they also saw a change in revenue (VSALES) positive, at 5.6% per year-with high dispersion, however. Average indebtedness (LEV) represented 29.2% of total assets and cash and cash equivalents (CX) 8.8%. It can be considered that the market value of the shares of these companies represented, on average, 2.34 times the book value of their equity in the period (MTB). The shares of these companies offered annual returns (RET) of 14.1% on average in the period, with a standard deviation (VOL) of 30%. The industry median indexes IMTB, IRET and IVOL were 1.39, 5.9% and 37.6% on average, respectively. The Herfindahl-Hirschman index (HHI) indicated low concentration in the industries on average (0.123), showing that markets were more competitive among publicly traded companies. The macroeconomic level proxy for Investment Opportunities (INVOP) is a standardized variable, however, it is possible to see that its average is greater than the median, indicating that the values to the right side of the distribution are further from the center. The average annual Selic rate in the period was 10.3%.

Table 2

Variable	N	Mean	Median	Std.Dev	CV	Minimum	Maximum
MA (t+1)	980	0.164	0	0.371	2.257	0.000	1.000
(Ln) EPU	980	5.144	4.987	0.441	0.086	4.347	5.726
(Ln) IIE-Br	980	4.652	4.638	0.093	0.020	4.497	4.766
(Ln) SIZE	980	15.712	15.656	1.594	0.101	10.973	19.874
ROA	980	0.045	0.045	0.067	1.500	-0.221	0.231
VSALES	958	0.056	0.034	0.273	4.880	-0.631	1.569
LEV	977	0.292	0.296	0.165	0.565	0.000	0.686
CX	977	0.088	0.667	0.079	0.900	0.000	0.409
MTB	977	2.238	1.425	2.229	0.996	0.170	11.896
RET	968	0.141	0.059	0.482	3,417	-0.721	1.889
VOL	980	0.300	0.251	0.164	0.545	0.136	1.131
IMTB	980	1.399	1.313	0.601	0.430	0.208	4.905
IRET	980	0.059	0.012	0.256	4.352	-0.345	0.750
IVOL	980	0.376	0.361	0.077	0.203	0.241	0.699
HHI	980	0.123	0.062	0.125	1.014	0.041	0.683
INVOP	980	-0.191	-0.738	1.929	-10.096	-3.322	2.658
SELIC	980	0.103	0.099	0.023	0.228	0.064	0.140

Note: N: number of observations. Std.Dev: Standard Deviation. CV: Coefficient of Variation. Obs.: The correlation matrix between the variables used in the regressions and their analysis can be found in Appendix B of this document.

=Figures 2 and 3 show the evolution of the EPU and IIE-Br and the number of M&A announcements of the sample companies over the period. The 128 companies in the sample engaged, on average, in 31.6 M&A transactions per year during the period. It should be noted that a company can make more than one announcement per year, so this volume is not restricted to one per company.

Figure 2. Note: The weighted average EPU over the months of each year was considered. Source: Data available at SDC Platinum and policyuncertainty.com.

Figure 3. Note: The weighted average IIE-Br over the months of each year was considered. Source: Data available at SDC Platinum and policyuncertainty.com.

The volume of deals announced fell sharply between 2010 and 2014 and did not return to the initial level (51 announcements in 2010) until 2019. On the other hand, the weighted average EPU showed an upward trend until 2016. The IIE-Br was less dispersed over the period, but with a jump from 100 to 120, approximately, in 2015, the year in which the announcements of the companies in the sample started to increase. It is important to highlight that the M&A announcements of the companies in the sample followed a different movement from the aggregate number of announcements in Brazil (PricewaterhouseCoopers, 2021). According to the PwC consultancy, the period from 2015 to 2019 showed a decrease in the average of transactions compared to the period from 2010 to 2014. In the next section, empirical tests of the statistical relationship between these variables will be presented.

Table 3 presents the results of the logistic regression of the propensity to acquire as a function of political and economic uncertainty in Brazil, considering the companies in the selected sample. For the first specification (1), the results were consistent with the hypothesis that economic policy uncertainty reduces the propensity of acquiring firms to engage in M&A activities. The negative coefficient (-0.754) was statistically significant. Furthermore, increases in the size of assests, on cash levels, stock returns and on their volatility, increase the likelihood that the firm will announce an acquisition in the following year. These results are more consistent with the findings by Borthwick et al. (2020), replicating the model in Chinese firms, especially for the sign of the volatility coefficient of returns (VOL) which was also positive, contrary to the results by Bonaime et al. (2018) with American companies, for this variable. As in the study by Sha et al. (2020), the estimated coefficients for the variables ROA, MTB and LEV had no statistical significance.

The coefficient of the variable at the industry level HHI was negative and statistically significant. This indicates that firms in less concentrated (and more competitive) sectors are more likely to announce an acquisition. Excluding the intercept, the other coefficients of this estimation were not statistically significant. In the second specification (2), the coefficient for the uncertainty proxy IIE-Br was not statistically significant.

Table 3

	Dependent Variable MA (t+1)
	(1)			(2)
	(Ln) EPU	Standard Error	|z|	(Ln) IIE-Br	Standard Error	|z|
Uncertainty	-0.754	0.386	-1.96*	-0.263	1.153	-0.23
(Ln) SIZE	0.338	0.091	3.73***	0.337	0.088	3.82***
ROA	1.513	1.921	0.79	1.670	1.936	0.86
VSALES	0.262	0.459	0.57	0.338	0.438	0.77
LEV	-0.054	0.355	-0.15	-0.033	0.350	-0.09
CX	2.145	0.907	2.37**	2.160	0.884	2.44**
MTB	-0.053	0.042	-1.27	-0.052	0.042	-1.24
RET	0.453	0.269	1.69*	0.422	0.259	1.63
VOL	1.286	0.778	1.65*	1.260	0.766	1.65
IMTB	0.203	0.310	0.66	0.183	0.293	0.62
IRET	-0.181	0.678	-0.27	-0.411	0.704	-0.58
IVOL	-0.052	1.959	-0.03	0.086	2.645	0.03
HHI	-6.539	3.378	-1.94*	-6.330	3.253	-1.95*
INVOP	-0.092	0.058	-1.59	0.047	0.082	0.57
SELIC	0.088	2.138	0.04	-0.603	3.567	-0.17
Constant	-3.206	1.911	-1.68*	-5.769	5.808	-0.99
Obs.		943			943
Pseudo-R2		0.0704			0.0671

Note: The area under the ROC curve indicated that the models have acceptable discriminatory power (> 68%). The dependent variable MA (t+1) takes the value 1 if the firm announces at least one acquisition in the following year and 0 otherwise. All independent variables are measured in period t. The sample consists of 128 selected companies, listed on B3 from 2010 to 2018. Model (1) assumes the weighted average EPU as a proxy for uncertainty, while model (2) assumes the weighted average IIE-Br. The coefficients estimated from the arithmetic mean had no statistical significance in any of the estimations. Models include robust standard error estimates with year clustering criteria and receive dummy controls for industry fixed effects. ^*, ^** and ^*** indicate the significance level at 10%, 5% and 1%, respectively.

We observed that the estimated coefficients for the control variables at the industry level IMTB, IRET, and IVOL, of these companies did not show statistical significance. These variables were inserted into the model of Bonaime et al. (2018) considering the Fama and French (1997) industry classifications for 48 industries with 115 thousand company year-observations. The limitation of the diversity of companies and industries that can be studied, considering companies listed in Brazil, and especially for this sample cut, may have contributed to the irrelevance of these variables in the present model. Moreover, Borthwick et al. (2020) replicating the model by Bonaime et al. (2018), also obtained results with low consistency for these variables, which may indicate that these variables in fact have low responsiveness for this model.

Nguyen and Phan (2017) and Sha et al. (2020) used more parsimonious models to try to explain the propensity for M&A. The common variables in at least 3 of the 4 studies cited are usually proxies for the size (Ln SIZE), cash (CX), profitability (ROA), sales variation (VSALES), leverage (LEV), market-to-book (MTB) and stock returns (RET). In particular, the stock return variable, which has a positive and significant estimated coefficient in all models, is consistent with the proposition by Harford (2005), on behavioral theoretical foundations that explain M&A waves. He explains that these waves coincide with bull market moments and are, consequently, positively correlated with the stock price and that this engages acquiring firms in M&A activities, who can also choose to pay in stocks, since they are overvalued. The fact that the sample consisted of listed companies may have contributed to this result. New studies in Brazil may seek to assess this relationship.

Having exposed the limitations on the data used in this research, we decided to estimate models with reduced specification, applying a filter to the models in Table 4, which is close to the proposal by Nguyen and Phan (2017) and Sha et al. (2020), including only variables that intersect between the models of these authors and the models by Bonaime et al. (2018) and Borthwick et al. (2020). Thus, the variables VOL, IMTB, IRET, IVOL and SELIC were excluded from the specification. The variables HHI and INVOP were maintained, as representatives of industry and macroeconomic control, respectively.

Table 4

	Dependent Variable MA (t+1)
	(3)			(4)
	(Ln) EPU	Standard Error	|z|	(Ln) IIE-Br	Standard Error	|z|
Uncertainty	-0.745	0.298	-2.5**	-0.352	0.800	-0.44
(Ln) SIZE	0.283	0.072	3.93***	0.283	0.071	3.97***
ROA	0.714	1.589	0.45	0.983	1.509	0.65
VSALES	0.300	0.428	0.7	0.383	0.398	0.96
LEV	-0.065	0.343	-0.19	-0.037	0.334	-0.11
CX	1.905	0.765	2.49**	1.901	0.760	2.5**
MTB	-0.049	0.047	-1.06	-0.049	0.046	-1.06
RET	0.457	0.183	2.5**	0.365	0.156	2.34**
HHI	-6.429	3.118	-2.06**	-6.087	3.168	-1.92*
INVOP	-0.090	0.046	-1.95**	0.045	0.072	0.63
Constant	-1.719	1.432	-1.2	-3.911	4.153	-0.94
Obs.		943			943
Pseudo-R2		0.0656			0.0619

Note: The area under the ROC curve indicated that the models have acceptable discriminatory power (> 67%). The dependent variable MA (t+1) receives the value 1 if the firm announces at least one acquisition in the following year and 0 otherwise. All independent variables are measured in period t. The sample consists of 128 selected companies, listed on B3 from 2010 to 2018. Model (3) assumes the weighted average EPU as a proxy for uncertainty, while model (4) assumes the weighted average IIE-Br. The coefficient estimated from the arithmetic mean was significant at the 5% level for estimation 3 and not significant in estimation 4. Models include robust standard error estimates with year clustering criteria and receive dummy controls for industry fixed effects. ^*, ^** and ^*** indicate the significance level at 10%, 5% and 1% respectively.

The smaller specification of the models would allow adopting a more rigorous level of significance (5%) for the interpretation and analysis of the results. Furthermore, the use of the arithmetic mean to represent the annual uncertainty index also produced significant statistics in the case of the EPU (estimation 3). In addition, the variable INVOP now has a negative and significant estimated coefficient. The signal obtained is consistent with the findings by Bonaime et al. (2018) and Borthwick et al. (2020). Despite the changes in its calculation for this study, the statistical significance and the verified sign validate this adaptation. This variable, formed from the extraction of the first principal component of the linear combination between macroeconomic variables, captures important properties of these variables and can avoid multicollinearity problems. Its use is suggested in other models that relate corporate decisions to macroeconomic factors.

The estimation of the logistic model using dummies for fixed sector effects can lead to a bias in the estimation, unless there are many companies per sector and the longitudinal dimension (T) of the panel is long (Wooldridge, 2019). The estimation of the marginal effect, which measures the effect of a unit change in the explanatory variable on the probability of occurrence of the event, could be harmed. In this sense, new models were estimated without industry fixed effects controls (Table 5).

Table 5

	Dependent Variable MA (t+1)
	(5)			(6)
	(Ln) EPU	Standard Error	|z|	(Ln) IIE-Br	Standard Error	|z|
Uncertainty	-0.638	0.300	-2.13**	-0.045	0.729	-0.06
(Ln) SIZE	0.252	0.042	6.03***	0.251	0.040	6.25***
ROA	0.756	1.586	0.48	0.975	1.497	0.65
VSALES	0.403	0.423	0.95	0.474	0.400	1.19
LEV	0.322	0.235	1.37	0.340	0.233	1.46
CX	1.778	0.844	2.11**	1.785	0.850	2.1**
MTB	-0.023	0.044	-0.52	-0;021	0.044	-0.49
RET	0.412	0.159	2.59***	0;320	0.129	2.49**
HHI	0.126	0.737	0.17	0;112	0.737	0.15
INVOP	-0.069	0.046	-1.48	0;054	0.069	0.79
Constant	-2.747	1.500	-1.83*	-5;780	3.475	-1.66*
Obs.		943			943
Pseudo-R2		0.0384			0.035

Note: The area under the ROC curve indicated that the models have acceptable discriminatory power (> 63%). The dependent variable MA (t+1) receives the value 1 if the firm announces at least one acquisition in the following year and 0 otherwise. All independent variables are measured in period t. The sample consists of 128 selected companies, listed on the B3 from 2010 to 2018. Model (3) assumes the weighted average EPU as a proxy for uncertainty, while model (4) assumes the weighted average IIE-Br. The coefficient estimated from the arithmetic mean was significant at the 5% level for estimation 5 and not significant in estimation 6. Models include robust standard error estimates with clustering criteria by year. *, ** and *** indicate the significance level at 10%, 5% and 1% respectively.

The marginal effect associated with the estimated coefficient for the EPU (-0.638), suggests that, cetris paribus, 1 unit increase is associated with a 0.0824 percentage point decrease in probability for mergers and acquisitions, given the unconditional probability of the announcement of an acquisition of 15.24%.

Table 6 summarizes the main results of the studies that related EPU and M&A, compared to the findings of this research.

Table 6

Models	Bonaime et al. (2018)	Borthwick et al. (2020)	Nguyen and Phan (2017)	Sha et al. (2020)	Model 1	Model 3
Country	USA	China	USA	China	Brazil	Brazil
Obs.	115,796	20,966	88,768	29,588	943	943
Pseudo-R2	-	0.02	0.07	0.01	0.07	0.07
Variables
Economic Policy Uncertainty	-	-	-	+	-	-
Size	+	+	+	+	+	+
Return on Assets	+	+	n/a	.	.	.
Sales Variation	+	+	+	n/a	.	.
Leverage	-	+	-	.	.	.
Cash and Equivalents	+	+	n/a	+	+	+
Operating Working Capital	n/a	n/a	+	n/a	n/a	n/a
Market-to-book	+	-	+	.	.	.
Past Returns	+	+	+	+	+	+
Returns Volatility	-	+	n/a	n/a	+	n/a
Shiller’s CAPE	+	.	n/a	n/a	n/a	n/a
Market-to-book (industry median)	.	+	n/a	n/a	.	n/a
Past returns (industry median)	+	.	n/a	n/a	.	n/a
Volatility (industry median)	-	-	n/a	n/a	.	n/a
Industry Economic Shocks	+	-	n/a	n/a	n/a	n/a
Herfindahl-Hirschman	n/a	n/a	n/a	n/a	-	-
Investment Opportunities	-	-	n/a	n/a	.	-
Interest Rate	+	+	n/a	n/a	.	n/a
Macroeconomic Uncertainty	-	+	n/a	n/a	n/a	n/a
Firm Age	n/a	n/a	+	n/a	n/a	n/a
Constant	.	-	-	-	-	.

Note: “+”for positive and significant coefficient; “-”for negative and significant coefficient; “.”For non-significant coefficient; n/a (does not apply) the model did not include the variable. All models have controls for industry fixed effects.

In Table 6 we include in the comparison only the estimated models that used the EPU variable (1 and 3), because it is the dimension of uncertainty analyzed in international studies. The results are consistent with international evidence, despite the considerably smaller number of observations, a fact that hampered the comparability of the studies.