This study looks at the effect of marketing and R&D investments on firm risk, using credit rating to measure firm risk. Data from 3905 COMPUSTAT firms are used for this study. Both logistic regression and machine learning techniques investigate the relationships between marketing and R&D investments and credit rating in this study. The results show that marketing investments are useful in reducing firm risk as it positively impacts a firms' credit rating.

Besides, firms operating in highly competitive markets should not fear competition. The results indicate a moderating effect of industry size on the relationship between marketing investments and firm risk. This moderating effect is also found for the relationship between R&D investments and firm risk. Both marketing and R&D investments positively affect credit rating when there is a lot of competition in the market. Moreover, this study also implies that through the Hidden Markov Model can be discovered what affects a positive or negative credit rating and the probability of going from a positive to a negative credit rating.

Keywords: Credit Rating, Firm Risk, Marketing Investments, R&D Investments, Hidden Markov Model

(3)

Acknowledgement

In this acknowledgement, I would like to thank my supervisor, Dr. Bhattacharya, for his good support while writing my thesis. His feedback has always been very helpful. Despite the fact that it was a different way of working due to the coronavirus, I experienced my thesis time as very valuable.

I hope you enjoy reading it.

Juliet Knipmeijer January 2021

(4)

1. Introduction

Interest in credit ratings has grown rapidly in recent years. Before the financial crisis in 2008, credit ratings could not accurately predict firms’ default risk. On the contrary, after the crisis, credit rating performance has improved (De Haan, 2017). Firms in the US are owing to an outstanding debt of 9.3 trillion dollars at the moment (Vazza, Kraemer & Gunter, 2019). During the current Corona crisis, firms will incur even more debt, but the chance that firms will default is high. Understanding credit rating is, therefore, even more, important to understand now.

Credit ratings estimate the vulnerability of a firm's cash flow and are essential because they show the link between return and risk (Rego, Billett & Morgan, 2009; Kumar & Rao, 2012). Since credit ratings are related to a firm’s risk, it is important for many parties such as firms, shareholders, and banks. For example, most commercial banks' capital is used for investments involving credit risks (Almansour, 2015). That is why research needs to gain more insight into what affects credit ratings.

There are several different ways of how firms can increase the value of investors in their firms. For example, firms can increase their cash flow, realize their cash flow earlier, extend their cash flow duration, and reduce their cash flow risk. Hence, investors such as shareholders and banks do not care only about the firm's expected returns, but they also care about a firm's risk. Investors care about firm risk because it is a way of assessing firms to which they can borrow safely, and they use credit ratings when determining the expected return rate based on the level of risk (Rego et al., 2009, Kumar & Rao, 2012). Of course, firms do even care about their risk, as less firm risk means higher loans, leading to riskier investments. Thus, firms need to have a high credit rating because this means that the firm has a low risk of default.

Within the marketing literature, not much has been looked into the effect of marketing and R&D investments on credit rating. There is literature that looked at the relationship between marketing and R&D investments and firm risk. Research showed, for example, that marketing investments could increase returns, but that also lowers the firm risk (Srivastava, Shervani & Fahey, 1997; McAlister, Srinivasan & Kim, 2007). Moreover, literature also showed that R&D investments are risky and associated with higher firm risk (Mata & Woerter, 2013; Zhang, 2015). So, while different studies already looked at the relationship between marketing and R&D and firm risk, they do not use credit rating to measure firm risk. Other types of risk, such as systematic risk and idiosyncratic risk, are interrelated to credit risk, but these risks are not the same (Haq & Heaney, 2012). Credit rating, which is related to credit risk, is important because credit rating is the default risk, unlike other types of risk (Ratings,

(6)

2016). The study of Rego et al. (2009) looked already at the effect of marketing on credit rating, but they used brand equity instead of marketing investments.

Therefore, in this study, I focus on the effect of marketing and R&D investments in explaining firm risk, using credit ratings to measure firm risk. Besides, I use competition as a moderator in these relationships. This moderator has never been examined in other studies related to marketing and R&D investments and credit rating. The following research question will be answered in this study: How do marketing and R&D investments influence firm risk?

After analyzing data from 3905 firms through COMPUSTAT, this study proposes several contributions to the literature. Firstly, this research implies that marketing and R&D are not always complementary, at least in credit rating. I found that marketing investments have a significant positive impact on credit rating and thus reduces firm risk. This effect also holds when the competition is fierce. Moreover, marketing investments lead to a positive change in credit rating. I also found that R&D investments significantly negatively affect credit rating, which means it increases firm risk. However, when the competition is fierce, R&D investments positively affect credit rating and reduce firm risk. Besides, exciting results are found through the Hidden Markov Model. This model showed the probability that a firm belongs to a specific state and describes the states. In this study, state one can be seen as the most beneficial state for firms since it is related to a positive credit rating. Moreover, the chance to go to a state with a negative credit rating is really low.

Webster, Malter & Ganesan (2005) showed that the marketing discipline is damaged over the past two decades by the rapid globalization of business and competitive pressures.

That is why firms are cutting back on marketing investments. However, this research shows three important managerial implications, where the first implication is related to marketing investments. This study shows that marketing investments are useful to reduce firm risk.

Therefore, managers should invest more in marketing to reduce firm risk. Moreover, this study also shows that managers should not fear competition, as the effect on credit rating is still positive when the competition is included.

This study's second implication shows that managers should only invest in R&D when competition in a market is fierce, as R&D without competition does not affect credit rating.

Moreover, since the combination of marketing and R&D investments does not affect credit rating, managers should not focus on both at the same time.

The final implication of this study is for investors. This research shows that investors should focus on smaller, healthy firms with a low market share that operate in a market where

(7)

competition is low. They should also target firms that invest a lot in marketing. Those firms are more likely to repay their debts, as these firms have a positive credit rating.

The remainder of this thesis is structured in the following way. Chapter 2 describes the literature review on firm risk combined with the hypotheses of this study. Chapter 3 explains the descriptives of the dataset and the methods used, combined with the conceptual model.

Moreover, chapter 4 will discuss the results of this study. Then, chapter 5 describes the discussion. Chapter 6 will discuss the implications of this study. Lastly, the limitations and advice for future research will be described in chapter 7.

(8)

2. Literature review

The literature review of this study is presented in this chapter. Next to that, the hypotheses formulated after reviewing the literature are also discussed in this chapter.

2.1 Firm risk

According to Rego et al. (2009), risk is essential in various areas, such as finance, marketing, accounting, and strategic management. The most investigated type of risk used in marketing is a variability-based risk. Variability in firms' cash flows leads to uncertain outputs because it is difficult to predict what will happen. Therefore, investors need higher returns to offset the lower predictability, leading to lower stock prices and higher debt. Another type of risk is the vulnerability risk. Vulnerability risk pertains to firms' cash flows measured as regards to the probability that the firm will meet their financial needs and obligations (Rego et al., 2009).

Ruefli, Collins & Lacugna (1999) showed that interest in risk at the firm level could differ for different stakeholders. For example, employees and investors can both have a different view of risk at the firm level. Both equity-holder risk and debt-holder risk significantly impact firm’s capital costs (Anderson, Mansi & Reeb, 2004). From the equity-holders' view, the risk is driven by the pricing model for capital assets. This model considers total equity risk related to the firm's stock returns variability (Rego et al., 2009). However, for debt-holders, the most crucial aspect of risk is the firm's future cash flows' vulnerability, which determines whether a firm can service its existing debt and its ability to incur and pay off the new debt.

This type of risk is also called credit risk (Rego et al., 2009). Credit risk relates to credit ratings, which measure firm risk in this study.

A credit rating is an expectation of the firm to repay its debt or loans based on a firm's creditworthiness (Afonso, 2003). A good rating helps firms get larger loans to expand or implement risky innovations. A rating agency's evaluation is done by a rating agency, which looks at the probability of default. This rating agency uses qualitative and quantitative information to define this credit rating. Factors affecting the credit rating include the likelihood of default, the prospect of recovery from the default, and credit stability (Ratings, 2016).

Heins, Leach & Mcgrath (2007) argue that intangible investments are critical drivers of competitive advantage and shareholder value, and therefore, investments are essential in credit ratings. According to Hilscher & Wilson (2017), firms with a lower credit rating are usually more likely not to survive than firms with a higher credit rating. Therefore, usually, firms with a lower rating have a higher systematic risk.

(9)

The study of Singal (2013) has shown that a credit rating measures the solvency of a firm, but that it is also related to the past, current and expected performance of a firm. Credit ratings are related to firm risk since they estimate the vulnerability of a firm's cash flow (Rego et al., 2009). Credit ratings are essential to firms since investors will use them to determine the expected return rate based on the risk level. When a firm's credit rating is low, it means the company has a high risk of default. Thus, a credit rating shows the link between return and risk (Kumar & Rao, 2012).

According to Gruca & Rego (2005), future cash flows should be maximized to minimize a firm's risk and maximize shareholders' value. Cash flows are a premature measure of firm performance and determine its value to its shareholders (Dechow, 1994). Stahl, Matzler

& Hinterhuber (2003) have shown that shareholder value is driven by increasing cash flows and reducing cash flows' volatility and vulnerability. According to Stahl et al. (2003), reducing volatility and vulnerability is important because higher volatility and vulnerability carry higher risks associated with cash flow streams, leading to a higher discount rate, resulting in a lower credit rating.

Several studies have shown that different types of variables are significant financial predictors of firm risk; leverage, liquidity, profitability, and retained earnings (Patel, Evans &

Burnett, 1998; Ferreira & Laux, 2007; Almansour, 2015). Therefore, standard financial models are replicated in this study using economic predictors to explain the firm risk. Afterward, the effect of the variables of interest (marketing and R&D investments) on firm risk are explored.

2.2 Marketing and R&D investments and firm risk

In the previous paragraph, I have shown which factors influence firm risk. This section argues that marketing and R&D investments influence the levels, volatility, and vulnerability of cash flows. These intermediates affect credit ratings since cash flows directly impact its ability to pay off its debt. Several studies have looked at the relationship between marketing and R&D investments with firm risk (Srivastava et al., 1997; Zhang, 2015). Still, extant studies do not use credit ratings as a measure of firm risk. Therefore, recent studies do not capture an essential aspect of firm risk since credit ratings are most important for bondholders and financial institutions.

2.2.1 Marketing investments and firm risk

Marketing investments are expressed as advertising investments in this study, as this is a form of marketing (Lilien & Weinstein, 1984). Srivastava et al. (1997) have shown that marketing

(10)

enhances shareholder value since it reduces the volatility and vulnerability of cash flows, reducing a firm's risk. This is also confirmed by the study of Chauvin & Hirschey (1993), which have found a positive effect of marketing on future cash flows. When the level of a firm’s cash flows is high, the firm risk decreases (Amit & Wernerfelt, 1990). However, Roumani, Nwankpa & Roumani (2016) have shown that marketing investments positively correlate with vulnerabilities. They argue that as firms invest more in marketing, the products' popularity increases, leading to greater product exposure. This can lead to more vulnerabilities and attacks.

The study by Lin, Lee & Hung (2006) has shown that marketing investments are essential for improving the long-term financial firm performance. Tobin's Q is their firm performance measure, which captures a firm's risk and returns dimensions. This is also confirmed by Konak's (2015) study, which shows that marketing has a significant positive effect on firm performance, which also uses Tobin's q to measure firm performance. According to McAlister, Srinivasan & Kim (2007), marketing investments create intangible assets shielded from changes in the stock market and reduce a firm risk.

Therefore, we expect the following:

H1: Marketing investments are positively associated with a reduction of firm risk

2.2.2 R&D investments and firm risk

R&D investments increase technical and scientific knowledge and apply that knowledge to improve products and processes (Hagedoorn, 2002). According to Mata & Woerter (2013), R&D investments are risky because firms' benefits from R&D investments vary. Moreover, Czarnitzki & Kraft (2004) also have shown that too much R&D investments may harm a credit rating. The reason for this is that R&D investments are always subject to possible failures.

Moreover, this is also confirmed by the study of Zhang (2015), which stated that R&D investments are inflexible and associated with high adaptation costs. They also found that firms investing in R&D may have severe financial constraints and are more likely to suspend R&D projects. Thus, R&D investments can lead to an increase in the distress risk of a firm.

H2: R&D investments are negatively associated with a reduction of firm risk

(11)

2.3 The effect of competition on the relation between marketing and firm risk

As already described earlier, debt, cash flows, and cash flow variance can be seen as firm risk drivers (Gruca & Rego, 2005, Rego et al., 2009). The study of Morellec, Nikolov & Zucchi (2014) has shown that the intensity of competition and cash flows are positively related. Thus, they argue that more competition leads to higher cash flows.

The study by Rust, Ambler, Carpenter, Kumar & Srivastava (2004) has shown that competition affects marketing investments. Under conditions of low competition, marketing investments can be more effective as there is less clutter (Henke, 2013). Besides, marketing investments reduces the likelihood of switching to competitors as marketing increases consumer loyalty and inertia (Carpenter & Lehmann, 1985). Furthermore, there may be less variance in cash flows as there are few alternatives that customers can choose from (and hence they have to stay loyal). When the competition is low, switching costs are high, and consumers are less likely to switch to an alternative, which will keep them loyal (Lee, Lee & Feick, 2001).

However, through inducing greater satisfaction and then loyalty, marketing can also ensure more 'true' commitment such that customers do not defect when presented with an alternative (Yang & Peterson, 2004). Hence, debtholders who provide service to firms that invest in marketing are more likely to be compensated for their debt without fail under conditions of low competition.

Conversely, at high levels of competition, marketing investments may be less effective, as the impact of competition on customer loyalty is high as consumers have more alternatives to choose from (Chen, 2015). Switching costs become more important when there are alternatives in the market. When switching costs are low, i.e., when the competition is fierce, consumers who are satisfied but not loyal are more likely to switch to alternatives. Moreover, firms operating in a highly competitive market also have higher bank debt costs (Valta, 2012).

Thus, debtholders who provide service to firms investing in marketing have a lower chance of being compensated for their debt without fail under conditions of high competition.

H3: The effect of marketing investments on reducing firm risk strengthens when the competition is low within the industry

2.4 The effect of competition on the relation between R&D and firm risk

The research by Woerter (2014) showed that competition influences the R&D investments of firms. They argue that when competing in a market is low, R&D investments are more

(12)

effective, as firms can better finance their R&D investments. Thus, debtholders who provide service to firms that invest in R&D are more likely to be compensated for their debt without fail under conditions of low competition.

Conversely, when competition in a market is high, Woerter (2014) found that R&D investments are less useful because firms have problems financing their R&D investments.

These firms are not rewarded for their incentive, as many other alternatives exist that a consumer can choose from (Martin, 1993). Moreover, the chance that R&D investments fail is higher in markets where the competition is high since rival firms could win the innovation race.

Thus, potential future cash flows associated with R&D investments have a higher chance of being extinguished when the competition is high (Gu, 2016). So, debt-holders who provide service to firms investing in R&D have a lower chance of being compensated for their debt without fail under conditions of high competition.

H4: The effect of R&D investments on reducing firm risk strengthens when the competition is low within the industry

2.5 The effect of the combination of R&D and Marketing on firm risk

The study by Leenders & Wierenga (2002) showed that when marketing and R&D are integrated, firms will achieve better performance of new products. More integration of marketing and R&D means that new products can be developed better and faster, with higher profits and lower costs (Griffin & Hauser, 1996). This means that the performance of a firm will increase, and firm risk will decrease. According to Feng, Morgan & Rego (2017), R&D capabilities positively affect the impact of marketing capabilities on firm growth, which is an important aspect of firm performance. Capabilities provide advantages in economies of scope for the firm’s investments in their knowledge resources (Morgan, Vorhies & Mason, 2009).

Given the findings with capabilities, investments may act similarly (Barney, 1991). Therefore, we expect that R&D investments also positively affect marketing investments. So, debtholders who provide service to firms combining R&D and marketing investments have a greater chance of being compensated.

H⁵: The combination of R&D and Marketing investments are associated with a reduction of firm risk

(13)

3. Method

The method of this study is presented in this chapter. First, some descriptions of the dataset are explained to get an idea of the data. This chapter also describes how the data is cleaned up and why. Then the measure used for this study is defined. After the measurement, the control variables used in this study are explained. Furthermore, the model estimation and specification will be described.

3.1 Descriptives

The data used for this study ranged from 1981 to 2017. The data includes information about 3905 firms obtained from COMPUSTAT. Descriptive statistics of the variables are presented in Table 1. The dataset consists of financial performance measures such as retained earnings, market share, liquidity, profitability, and leverage. Besides, marketing and R&D variables are used from this dataset, such as marketing and R&D investments. The dependent variable called credit rating is used to see whether a firm's rating is positive or negative, and the predictive variables' effect on a positive/negative credit rating. A moderator called industry size is created to see whether competition also affects credit rating.

The dataset includes 45003 positive credit ratings (AAA till BBB+). On the other hand, the dataset includes 1424 negative credit ratings (C till D). A total of 13,938 changes occurred in the dataset, with 7,303 negative changes and 6,635 positive changes. Most observations remain the same, namely, three-quarters of the observations. The number of firms that change to a better rating is 3036, while 3118 firms change to a lower rating. On average, it takes 2,73 years for firms to switch to a better rating.

Furthermore, the average sales amount for the firms in the dataset is 8494.0 million.

Finally, an additional variable called rank is created based on the firms' lagged ranks to see the predictive variables' effect on the change in credit rating (1 till 22). Where rank 22 relates to credit rating AAA, rank 1 relates to credit rating D.

(Insert Table 1)

3.2 Data cleaning

To account for anomalies in the dataset, the data must be cleaned up before it can be estimated.

Some missing values have been found in the dependent variable credit rating; these missing values have been deleted list-wise. The variables R&D capability, marketing capability, goodwill, market share, profitability, marketing investments, liquidity, and leverage have most

(14)

of the missing values. Since the dataset is large, list-wise deletion should not affect the analysis outcome (Allison, 2000). Kang's study (2013) has shown that when the sample is large enough and power is not a problem, list-wise deletion is a good strategy. Since the percentage of missing data is relatively small, and the sample is large, there is still enough power to find stimulating effects.

The predictive variables also contained missing values; these missing values have been deleted case-wise. Also, some extreme outliers are found in the dataset. According to Aguinis, Gottfredson & Joo (2013), outliers can be seen as a unique phenomenon that can lead to new exciting insights. Moreover, outliers often lead to significant changes in substantive conclusions. Removing outliers can lead to an incorrect acceptance or rejection of the hypotheses (Aguinis et al., 2013). That is why these outliers are included in the dataset. To be able to compare the coefficients of the different variables, variables are standardized.

Standardization makes it easier to compare the different variables' importance because they have the same scale (Bring, 1994).

The correlation matrix (Table 2) shows that the variable marketing capability correlates with the variables retained earnings (0.91), marketing investments (0.90), R&D investments (0.78), and industry size (0.72). Therefore, the variable marketing capability will not be used for further analysis. The variable R&D capability also shows a significant correlation with marketing investments (0.44) and a moderately significant correlation with profitability (0.10).

So, the variable R&D capability will also not be used for further analysis. Lastly, goodwill and R&D investments are moderately significantly correlated (0.11). Therefore, the variable goodwill will also be deleted from the dataset. Lastly, leverage and industry size are moderately significantly correlated (0.17) but will be used for further analysis. Table 3 shows that all VIF scores of the different variables, including leverage and industry size, are less than 4. Therefore, it is assumed that there is no multicollinearity in this model, and therefore all variables will be used for further analysis.

According to Cain, Zhang & Yuan (2016), skewness is a measure that can be used to understand normality in the data. If skewness differs from 0, the distribution diverges from symmetry. Skewness is expected to be 0 for a distribution that is symmetrical, like a normal distribution. The study of Cain et al. (2016) states that: ‘Distributions with positive skewness have a longer right tail in the positive direction, and those with negative skewness have a longer left tail in the negative direction’ (p.1717).

(15)

Table 4 shows that all variables are slightly positively skewed. According to George &

Mallery (2010), the value of skewness should be between -2 and +2. Therefore, the values of skewness in this dataset can be considered as acceptable.

(Insert Table 2, 3 and 4)

3.4 Dependent and independent variables

Credit rating. The dependent variable used in this study is credit rating. This variable is based on Compustat's Standard & Poor's Issuer Credit Rating (ICR). It looks at the overall creditworthiness of firms (Ratings, 2016). The variable credit rating in this study contains 22 credit ratings ranging from AAA to D. Where AAA is the highest credit rating a firm could have and D the lowest. The variable in this study has been split up between BBB + and C. A credit rating between AAA and BBB+ means a positive credit rating. From C onwards, the credit rating becomes negative. The dependent variable is a dummy variable where 1 means a positive credit rating and 0 means a negative credit rating.

Marketing investments. The variable marketing investments is based on advertising media costs. This could be for example radio, television, periodicals and promotional investments (Services, n.d.). The units are expressed in millions.

R&D investments. The variable R&D investments is based on the costs that are related to developing new products or services during the year (Services, n.d.). The units are expressed in millions.

Industry size. The variable industry size is based on the number of unique company keys within an industry based on the sic industry codes. The Herfindahl Hirschman Index (HHI) is used to understand the level of competition within an industry (Pavic, Galetic &

Piplica, 2016). Figure 1 shows that some industries are highly concentrated, but many industries are highly competitive. Therefore, the variable industry size can be used to measure competition (Abdoh & Varela, 2018).

(Insert Figure 1) 3.5 Control variables

Control variables are added to the model in this study to determine whether the independent focal variables behave as assumed (Nielsen & Raswant, 2018). Since credit rating is based on different firm performance indicators (Samson, 2006), this study will use control variables that

(16)

also influence a firm's performance. This study's control variables are market share, liquidity, leverage, profitability, and retained earnings.

Market share. Research by Kim & Mauborgne (2005) has shown that a market expansion strategy can be seen as an alternative to a market share growth strategy. Moreover, Bang & Joshi (2010) shows that a market expansion strategy is positively correlated with a firm's sales revenue and profits.

Liquidity. According to Fang, Noe & Tice (2009), there is a positive relationship between liquidity and firm performance. This study has shown that the higher the firm's liquidity, the better its performance. The reason for this is that the more liquidity a firm has, the more access a firm has to the information content of market prices. Moreover, liquidity leads to performance-sensitive managerial compensation (Fang et al., 2009).

Leverage. Leverage has a significant effect on the performance of firms. The study of González (2013) shows that firms with higher leverage appear to have a reduced effect on firm performance. This is also confirmed by Ibhagui & Olokoyo (2018) study, which has shown that leverage harms firm performance with a small size. According to Vithessonthi & Tongurai (2014), leverage positively affects firm performance when it is international-oriented.

However, leverage has a negative effect on firm performance when the firm is domestic- oriented.

Profitability. The study of Chen & Chen (2011) has shown that profitability positively affects a firm's value. The more profits a firm has, the better it can distribute earnings for its shareholders, and therefore the expected value of a firm will be higher (Yang, Lee, Gu & Lee, 2010).

Retained earnings. According to Ball, Gerakos, Linnainmaa & Nikolaev (2020), retained earnings are a good predictor of a firm's returns. Moreover, they have shown that retained earnings are a good indicator of underlying earnings.

3.5 Conceptual model

The conceptual model in Figure 2 shows the relationships and hypotheses proposed in this study. As described before, credit rating will be used as a measure of firm risk in this study.

(17)

Figure 2: Conceptual model

3.6 Model Specification

This study's model tests the effect of marketing and R&D investments on credit rating, a measure of firm risk. A full model with interactions is used in this study. First, a model is tested with only control variables. A model is then tested with control variables and the independent variables to see the effect of marketing and R&D investments on credit rating. Finally, a model is tested with interaction effects. Together with the control variables, the following equation is provided:

Credit rating = 𝛼 + 𝛽1MarketingInvestments + 𝛽2R&DInvestments + 𝛽3RetainedEarnings + 𝛽4Marketshare + 𝛽5Profitability + 𝛽6Liquidity + 𝛽7Leverage + 𝛽8Industrysize + 𝛽9MarketingInvestments*Industrysize + 𝛽10R&DInvestments*Industrysize ⁺ 𝛽11MarketingInvestments*R&DInvestments

3.7 Model Estimation

Logistic regression. Various logistic regression models are first performed for the estimation of the models. Logistic regression models are used to estimate models where the dependent variable is binomial (Oshagbemi & Hickson, 2003). Three different logistic regression models are performed to see if there is an effect of marketing and R&D investments on credit rating and whether there are interaction effects on credit rating.

Machine learning. Two other machine learning methods are used and compared:

random forest and boosting. Machine learning models are used since they predict better than

(18)

logit models, and the interpretation for machine learning models is more straightforward.

Compared to logit models, machine learning is based on regularization and variable selection (Phan & Thi, 2019). Also, logit models are more likely to overfit. Thus, because the predictive accuracy for random forest models is higher than for logit models, a random forest model will be used and compared (Zhao, Yan, Yu & Hentenryck, 2020). Besides, the data in this study include rare events. Since boosting is more effective in predicting rare events, the boosting method will also be used and compared (Blagus & Lusa, 2017).

To compare the different models' predictive performance, measures such as Top- Decile-Lift, GINI coefficient, and Hit Rate are used. These measures are most often used for data where the decision is binomial (De Haan, Verhoef & Wiesel, 2015). Top-Decile-Lift shows the power a model has to defeat the average performance or random model. Top-Decile- Lift uses the top 10% of respondents to measure predictive performance (Xie, Li, Ngai & Ying, 2009). According to Bendel, Higgins, Teberg & Pyke (1989), 'The GINI coefficient is often defined in reference to the Lorenz curve, which results from a plot of the cumulative proportion of the population to the cumulative proportion of the variable’ (p.395). The Hit Rate shows the percentage of observations that the model has correctly predicted. However, when the Hit Rate is high, it does not always mean that the model is good at predicting. The sample of binary choice models may have an uneven distribution between the two possible outcomes (Donkers

& Melenberg, 2002). Because the dependent variable is unevenly distributed, the SMOTE method is used for the boosting model. SMOTE is an approach for constructing classifications in unbalanced datasets (Chawla, Bowyer, Hall & Kegelmeyer, 2002).

Fixed effect analysis. Another popular statistical method will be performed to see whether the predictive variables influence the credit rating change. A fixed-effect analysis is performed to check for unobserved heterogeneity in panel data (Bhattacharya, Misra &

Sardashti, 2019). The Hausman test is part of the fixed effect analysis and shows whether to use a fixed effect or random effect analysis (Schmidheiny, 2019). According to Borenstein, Hedges, Higgins & Rothstein (2010), it is assumed for the fixed-effect model that one true effect size underlies all the studies in the analysis. Besides, it is also assumed that the differences that are observed are due to sampling error. In contrast, for the random effect analysis, it is assumed that the actual effect sizes differ, but the effect size may also differ from study to study (Borenstein et al., 2010).

Hidden Markov Model. According to Beal, Ghahramani & Rasmussen (2002), Hidden Markov models are often used methods to model sequences in machine learning. The study of Beal et al. (2002) states that 'An HMM defines a probability distribution over sequences of

(19)

observations by invoking another sequence of unobserved, or hidden, discrete state variables’

(p.1). Besides, the hidden Markov model tells us the optimal number of states and describes the different states. The states' description goes more in-depth as the Hidden Markov Model also shows us what the transition probabilities are, in other words, the chance to move from one state to another or remain in the same state (Beal et al., 2002). The Hidden Markov Model also shows us the emission probabilities. According to Jackson (2011), this is 'the probability density of the outcome conditional on the hidden state’ (p.13).

In this study, the Hidden Markov Model will be applied to see if there are hidden states between credit ratings. It is investigated whether the states depend on marketing and R&D variables. In other words, this study attempts to predict the next credit rating, or change in credit rating, based on different variables in the data set using the Hidden Markov model. This is interesting because changes in the credit ratings are uncommon, and there may be several states a company goes through to get the next credit rating.

Several measures, such as AIC and BIC and Log-Likelihood, are used to determine the optimal number of states (Celeux & Durand, 2008). Table 5 shows the Hidden Markov Models with two, three, four, and five states. According to Table 5, the Hidden Markov model with three states is the optimal model since this model has the lowest BIC and AIC compared to the other models. Therefore, this Hidden Markov model with three states will be used for further analysis.

(Insert Table 5)

(20)

4. Results

The results of the study are presented in this chapter. First, the results of the logit model and the machine learning models are described. Furthermore, a fixed-effect analysis is performed to see the predictive variables' effect on credit rating change. Finally, the results of the Hidden Markov model are described. The purpose of this chapter is to compare the results of the analysis with the established hypotheses.

4.1 Empirical results

As explained before in the method, multicollinearity and correlation between the variables are checked before explaining the different models that test this study's hypotheses.

Furthermore, the results of the machine learning models, fixed-effect analysis, and the Hidden Markov model are described in this section.

4.1.2 Model fit

To compare the different logit models that are used, measures as AIC and BIC are used.

As explained before in the method section, variables are standardized to compare the different variables' coefficients. The results in Table 6 show that the logit model with marketing and R&D variables performs better than without marketing and R&D variables since the AIC and BIC measures are lower (AIC; 542.61, BIC; 599.60). Moreover, the logit model with direct effects and interaction effects performs best (AIC; 540.47, BIC; 616.45). So, when marketing and R&D variables and interaction effects are included in the model, the model is better at predicting.

(Insert Table 6)

To compare the logit models' different performances with the random forest and boosting model, GINI and TDL measures are used. Table 7 shows that the logit models and the random forest model perform the same in terms of Hit Rate (0.98) and TDL (1.02) and almost the same in terms of GINI (0.06). However, the boosting model performs best in terms of GINI (0.11), TDL (1.03), and Hit Rate (100%). This relates to Blagus & Lusa’s (2017) study, which shows that boosting is more effective in predicting rare events.

(Insert Table 7)

(21)

4.1.3 Logit models

First, a logit model is performed without marketing and R&D variables. Table 8 shows that retained earnings (beta = 0.34, p <.001), profitability (beta = 0.56, p<.001), and liquidity (beta = 0.63, p<.001) have an immediate positive significant effect on credit rating being of 1 (i.e. having a positive credit rating). So, the probability of having a positive credit rating increases when a firm's retained earnings, profitability, or liquidity increase. Also, profitability and liquidity have more influence on having a positive credit rating than retained earnings. In contrast, leverage has a direct negative significant effect on credit rating being 1 (beta = -0.67, p <.001). So, when a firm's leverage increases, the probability of having a positive credit rating decreases. Finally, market share and the industry's size (competition) have no significant effect on credit rating being 1 (i.e., having a positive credit rating).

(Insert Table 8)

Second, a logit model is performed with marketing and R&D variables to see those direct effects of marketing and R&D on credit rating. Table 9 shows that marketing investments significantly affect credit rating being 1 (beta = 2.83, p<.05). So, when a firm uses more marketing, the probability of having a positive credit rating increases. Moreover, R&D investments significantly negatively affect credit rating being 1 (beta=-0.25, p<.01). This means that when a firm invests in R&D, the probability of having a positive credit rating decreases. Since the variables are standardized, we can compare the effects of the different variables. Table 9 shows that the effect of marketing is roughly eleven times larger than R&D.

(Insert Table 9)

Lastly, a logit model is performed with interaction effects. According to Table 10, the moderation effect of industry size with marketing investments is significant (beta=2.24, p<.05).

So, when the industry is bigger, marketing investments have more effect on having a positive credit rating. The moderation effect of industry size with R&D investments is also significant (beta=0.31, p<.1). Thus, when the industry is bigger, R&D investments have more effect on a positive credit rating. However, when the industry is bigger, marketing investments have more effect than R&D investments. Moreover, there is no effect on the interaction between marketing investments and R&D investments on credit rating.

(Insert Table 10)

(22)

4.1.4 Fixed effect analysis

As described in the method part, a Hausman test has been performed on the data to judge whether a fixed or random analysis should be used (Schmidheiny, 2019). The null hypothesis, in this case, is the random-effects model. According to the Hausman test in Table 11, a fixed effect analysis should be used (p<.05).

(Insert Table 11)

As shown in Table 12, marketing investments, market share, retained earnings, profitability, and liquidity lead to a higher rank. However, leverage leads to a lower rank. Moreover, R&D does not affect the change in rank.

Positive change in credit rating (higher rating)

Negative change in credit rating (lower rating)

Marketing investments

X

Market share X

Retained earnings X

Profitability X

Leverage X

Table 12: Fixed effect analysis

4.2 Hidden Markov Model

As described in the method, the Hidden Markov Model is performed to determine what the hidden states are (Beal et al., 2002). We will describe the hidden states through emission probabilities, transition probabilities, and predictive variables. The predictive variables used to describe the hidden states are marketing investments, R&D investments, market share, retained earnings, profitability, liquidity, leverage, and industry size. As described in the method, the Hidden Markov Model with three states is the optimal model.

(23)

4.2.1 Emission probabilities

To expand the description of the different states, emission probabilities are described in Table 13. Table 13 shows that observations in state one and state two are mostly firms with a positive credit rating. The density probability of having a positive credit rating in states one and two is greater than 1. So, this state is likely to represent a 1. Moreover, state three observations usually have a negative credit rating. The density probability of having a negative credit rating in state three is 99.9996%, so this state is likely to represent a 0.

(Insert Table 13)

4.2.2 Transition matrix

To see which state is followed by which state, transition probabilities are calculated (Table 14). Figure 3 shows that almost no transitions are observed. The probability that firms will change from state one to state two is only 26%. However, most often, firms will stay in the same state. The probability of going from a negative credit rating to a positive credit rating is very small (0.00). This means that it is not likely that a firm will change from a positive credit rating to a negative credit rating.

(Insert Table 14)

(Insert Figure 3) 4.2.3 Description of states

To describe the states through the predictive variables, posterior probabilities are calculated (Table 15, 16, 17). Tables 18 show the influence of the different predictive variables on the states.

Marketing investments

Market share

Retained earnings

Profitability Liquidity Leverage R&D investments

Industry size

State 1:

Healthy firms

+ - + + + - -

State 2:

Innovative, risky firms

- - - - + +

State 3:

Expelled firms

- + - + - +

Table 18: Description of states

(24)

Observations with many marketing investments are likely to happen in state one since the effect of marketing investments on state one is positive. Therefore, the probability of having a positive credit rating is higher for firms that invest a lot in marketing. The effect of marketing on state two and three is negative. Therefore, state one and state two, both characterized as a positive state, are different in marketing investments.

Moreover, market share has a negative effect on states one and two. Therefore, the probability of having a positive credit rating decreases when firms have more market share.

The effect of market share on state three is positive. So, firms with more market share relate to state three. The probability of having a negative rating increases when firms have more market share.

Retained earnings have a positive effect on state one but no effect on state two and three. So, the higher the firm's retained earnings are, the more likely it is that the firms belong to state one. In order words, the probability of having a positive credit rating increases for firms with high retained earnings. Again, states one and two are different in retained earnings since retained earnings positively affect state one but do not affect state two.

Profitability has a positive effect on state one. So, the higher the firm's profits, the more likely it is that a firm belongs to state one and the higher the probability of having a positive credit rating. In contrast, profitability has a negative effect on states two and three. So, firms with fewer profits are probably in states two and three.

Moreover, liquidity has a positive effect on state one, while it has a negative effect on state two. So, the more liquidity a firm has, the more likely it is that the firm belongs to state one. Therefore, the higher the firm’s liquidity, the higher the probability of having a positive credit rating. Again, states one and two differ in terms of liquidity. Moreover, liquidity does not affect state three.

Leverage has a negative effect on state one, but a positive effect on states two and three.

So, the higher the firm's leverage, the more likely it is that the firm belongs to state two or three. Thus, leverage can lead to a positive and negative credit rating.

Firms with a lot of R&D investments probably have many observations in state two, while firms that do not have a lot of R&D investments most likely have many observations in state three. So, the more R&D investments a firm has, the higher the probability of a positive credit rating. However, there is still a risk of ending up in a negative credit rating, as the transition probability from state two to state three is 0.05%. Moreover, R&D investments do not affect state one.

(25)

Lastly, industry size (competition) has a positive effect on state three. So, firms that operate in a bigger industry are more likely to belong to state three. This means that the bigger the industry is, the higher the probability of having a negative credit rating. Industry size has a negative effect on state one, so firms that operate in a smaller industry most likely belong to state one. Moreover, industry size does not affect state two.

(26)

5. Discussion

The results of this study presented in Chapter 5 are discussed in this section. Besides, the results are compared with the literature mentioned in this study. This study's research question was as follows: How do marketing and R&D investments influence firm risk?

The focus of this study is based on firm risk. Firm risk is important because banks and shareholders are concerned about firm risk. They want to find firms to which they can borrow safely (Kumar & Rao, 2012). Previous studies showed that marketing investments could influence firm risk (Srivastava, Shervani & Fahey, 1997; McAlister, Srinivasan & Kim, 2007).

However, these studies do not use credit ratings as a measure of firm risk. Rego et al. (2009) used credit ratings to measure firm risk, but they used a different marketing type, namely brand equity. Brand equity is a marketing asset, while in this study, marketing investments are used, which are marketing expenses. An asset is something the firm already has and is going to be less risky. However, assets, such as brand equity, take more time to change because it is a long- term concept (Huang & Sarigöllü, 2014). Marketing expenses generate positive returns in the short term (Murthi, Srinivasan & Kalyanaram, 1996). Furthermore, marketing expenses are more under managerial control. For example, when a brand loses market share, managers are in control and can increase marketing expenses (Jaworski, 1988). In this study, if the manager wants to lower the credit risk, increases marketing expenses is the first way to go.

Moreover, previous literature have shown that R&D investments increase firm risk (Mata & Woerter, 2013; Zhang, 2015), which is also confirmed by this study because R&D investments have a negative effect on credit rating and so increase firm risk. However, these studies do not use credit rating as a measure of firm risk. Czarnitzki & Kraft (2004) studied the relationship between R&D intensity and credit rating, but they did not use competition as a moderator in this relationship. Interestingly, however, in this study, the effect of R&D investments on firm risk turns into a positive effect under fierce competition. So, when there a lot of competition in the market, R&D investments decrease firm risk.

Five different hypotheses were formulated to answer the research question in this study.

The first hypothesis was related to the direct effect of marketing investments on firm risk.

Previous literature in this study showed that marketing investments have a positive effect on future cash flows and that they reduce firm risk. Since marketing reduces volatility and vulnerability of cash flows, it enhances shareholder value and reduces a firm's risk (Srivastava et al., 1997). Moreover, marketing investments create intangible assets protected from changes in the stock market and reduce a firm’s risk (McAlister et al., 2007). This study shows that

(27)

when firms invest more in marketing, the probability of having a positive credit rating increases, and thus firm risk decreases. This result is consistent with the effect that was assumed in the hypothesis.

The second hypothesis was related to the direct effect of R&D investments on firm risk.

As discussed in the literature, R&D investments are risky since the benefits a firm derives from these investments vary (Mata & Woerter, 2013). Moreover, R&D investments are always subject to possible failures since R&D investments are inflexible and associated with high adaption costs (Czarnitzki & Kraft, 2004; Zhang, 2015). This study indicates that R&D investments are negatively related to credit rating and do not reduce firm risk. This indicates that when firms invest more in R&D, the probability of a positive credit rating decreases, and therefore the firm risk increases. This result is consistent with the effect that was assumed in the hypothesis.

The third hypothesis proposed that marketing investments strengthen the reducing effect on firm risk when the competition is low within the industry. The results of this study show that this hypothesis is not confirmed. The results show that when the competition in a market is high, marketing investments have more effect on having a positive credit rating, and in turn, reduce firm risk. One explanation for this could be that firms will not get away if they do not invest in marketing in a competitive market. Heil & Robertson (1991) argued that it is important for firms to identify and respond to competitor behavior, such as marketing. If firms do not invest in marketing when the competition does, they will not survive. Also, firms that invest in marketing can grow and have more potential in the competitive market (Hunt &

Arnett, 2006). Moreover, it is also a sign that firms are healthy when they invest in marketing since firms show that they have money to spend on marketing.

The fourth hypothesis proposed that R&D investments strengthen the reducing effect on firm risk when the competition is low within the industry. This hypothesis is also not confirmed by this study. Thus, when the industry is bigger, R&D investments affect a positive credit rating and thus reduce firm risk. One explanation for this could be that firms have to be unique to stand out in a competitive market. Firms must be unique because if there are more products to choose from in a competitive market, consumers are more likely to find products that fit their ideal preferences (Cachon, Terwiesch & Xu, 2008). Investing in R&D can also be seen as a sign of health, as firms show that they have enough money to invest in R&D (Nicolau

& Santa-María, 2015). However, when the competition in a market is high, marketing investments have more effect than R&D investments.

(28)

The latter hypothesis suggested that the combination of investment in R&D and marketing reduces firm risk. This study does not confirm this hypothesis. This study shows that the combination of R&D and marketing do not affect firm risk. An explanation for this effect could be that the integration of marketing and R&D might be an issue. Gupta, Raj & Wilemon (1985) argued that firms often get excited about a new idea without paying enough attention to its commercial significance. Likewise, the firms misinterpret their customers' needs and waste their scarce resources on products that do not meet the actual market needs.

In addition to the hypotheses conducted in this study, some additional results are performed. We performed a fixed-effect analysis to determine the effect of marketing and R&D investments on credit rating change. The results of this analysis show that marketing investments lead to a positive change in credit rating. Moreover, R&D investments do not affect the change in credit rating.

As discussed earlier, in this study, the Hidden Markov Model is performed to investigate whether a firm belongs to a specific 'hidden' state to achieve a specific credit rating.

The results of this study show that the three states are the optimal number of states. State one and state two were related to firms with a positive credit rating. Whereas state three was related to firms with a negative credit rating. The transition probabilities show that almost no transitions were likely to be observed. The change from state one to state two shows a probability of 26%, whereas the change from state two to state three showed a probability of 0.05%. Thus, the probabilities that firms would change to another state are low. This means that most firms will stay in the same state.

State one includes firms with a smaller market share that invests a lot in marketing.

Additionally, in state one, firms have high retained earnings and profits. The liquidity of these firms is also high, while the leverage is low. The firms in state one operate in a market where competition is low. That is why these companies are referred to as "healthy firms." State two includes firms with a smaller market share that do not invest much in marketing. These firms have low profits and liquidity. Besides, the leverage of firms in state two is high, and these firms invest a lot in R&D. That is why these companies are called "innovative but risky firms."

Finally, state three includes firms with a higher market share that do not invest much in marketing. The profits of these firms are low, and the leverage of these firms is high. Firms in state three operate in a market where competition is fierce. Therefore these firms are referred to as "expelled firms."

Thus, state one can be considered the most favorable state for firms because it is related to firms with a positive credit rating. Besides, the probability of switching from state one to a

(29)

negative state is 0.00%. This shows that it is difficult to change from credit rating since credit ratings are very sticky. When a firm wants to change to another state, they have to do much better in many aspects, for example, in marketing investments, retained earnings, profitability, and R&D investments. This study shows that the HMM model is a good model for discovering what affects a positive or negative credit rating and the probability of going from a positive to a negative credit rating.

In conclusion, this study shows that marketing investments are indeed important to reduce firm risk. Moreover, the study confirms that R&D investments might increase firm risk.

This study also finds evidence that marketing and R&D investments are useful when the competition in a market is high. This is a new finding since most literature showed that marketing and R&D investments affect firm risk when the competition is low (Henke, 2013;

Lee et al., 2001; Woerter, 2014).

(30)

6. Implications

The implications of this study are discussed in this chapter. The findings of this study are interesting because firms, banks, and shareholders care about credit ratings. Credit ratings are important to firms as a higher credit rating gives them higher investment loans. Besides, higher credit ratings also lead to a lower debt ratio for firms. Banks and shareholders care about credit ratings because they use credit ratings as a tool to find firms to which they can borrow safely (Kumar & Rao, 2012). Credit ratings are important to banks and shareholders because they do not want to lend money to risky firms.

First of all, this study shows that firms need to invest in marketing to reduce firm risk, as this will bring firms high returns and lead to a positive change in credit rating. When the competition in a market is low, marketing investments increase firm risk. However, when the competition in a market is high, marketing investments reduce firm risk. So, when firms operate in highly competitive markets, they should invest in marketing and not fear competition. They still have to focus on empty spaces, as competition does have a negative effect on sales in general.

Besides, this study shows that R&D investments have a negative effect on firm risk and that R&D investments do not affect the change in credit rating. But interestingly, this study shows that R&D investments are super useful when competition in a market is fierce. On the contrary, R&D investments increase firm risk when there is little competition. Therefore, firms must invest in R&D when the competition is fierce. This finding is also interesting for firms that are good at R&D but that do not operate in a competitive market. Hence, when firms operate in highly competitive markets, they should invest in R&D and not fear competition.

They should still focus on empty spaces as competition negatively impacts sales overall.

However, while R&D investments are useful, marketing investments are more useful.

The effect size is almost eleven times larger. This study shows that marketing, combined with R&D investments, does not affect credit rating and, therefore, firm risk. Thus, firms should not invest in both. Hence, firms should focus more on marketing investments.

Moreover, this study shows that different types of firms could be categorized through the Hidden Markov models, which is very interesting. Such categorization has never been done before. According to the Hidden Markov model results in this study, investors can look at healthy firms that operate in a small market (low competition). Moreover, they can also look at firms that use a lot of marketing. Furthermore, investors should focus on smaller firms with

(31)

a lower market share. Those smaller firms have a higher probability of having a positive credit rating. An explanation for this could be that they have a higher chance of growth.

This study contributes to the literature as it shows that marketing and R&D are not always complementary, at least in terms of credit rating. This study shows that marketing investments have a significant positive effect on credit rating and thus reduce firm risk. If competition is fierce, this effect will continue to apply. This means that competition should not be a fear of firms. This study also shows that marketing investments can lead to a positive change in credit rating. Besides, this study shows that R&D investments have a negative impact on credit rating. But when the competition in a market is high, the effect of R&D investments changes into a positive effect on credit rating. Finally, this study shows that the Hidden Markov model can lead to interesting results to categorize firms. In this study, it is interesting to see what types of firms investors should or should not focus on.

(32)

7. Limitations and Future Research

In addition to the interesting implications this research has, this study also has some limitations.

In this section, the limitations of this study are described. Also, based on the limitations, suggestions for future research are given.

The first limitation of this study is related to the fact that all firms used for this study are publicly listed firms from Compustat (Ratings, 2016). In this study, we did not look at private firms. According to Hope, Thomas & Vyas (2013), financial reporting can differ for private firms. They showed that public firms are more conservative and that their accrual quality is higher than for private firms. Future research could explore whether marketing investments are also useful for private firms. Moreover, future research could also reveal whether the effect of R&D investments on credit rating is different for private firms.

The second limitation of this study is related to reverse causality. In this study, we did not test reverse causality. According to Leszczesky & Wolbring (2019), reverse causality may threaten the causal inference. In this study, we did not test whether firms with a high credit rating can do more marketing and thus get an even better credit rating or whether because of marketing, they get a better credit rating. In other words, in this study, it is not clear what came first, marketing or credit rating. Future research could test reverse causality between marketing investments and credit rating.

The last limitation of this study is related to the moderator industry size. In this study, we did not look at industry-wide differences outside of competition. Another interesting moderator could be, for example, digitalization within industries. Eling & Lehmann (2018) showed that new products could be developed through digitalization opportunities. Thus, one could expect that digitalization within the industry might have a positive impact on R&D investments. Therefore, for future research, it is interesting to investigate whether other industry moderators, such as digitalization within the industry, affect the relationship between R&D investments and credit rating.

(33)

References

Abdoh, H. A. A., & Varela, O. (2018). Product market competition, cash flow, and corporate investments. Managerial Finance.

Afonso, A. (2003). Understanding the determinants of sovereign debt ratings: Evidence for the two leading agencies. Journal of Economics and Finance, 27(1), 56-74.

Aguinis, H., Gottfredson, R. K. & Joo, H. (2013). Best-Practice Recommendations for Defining, Identifying, and Handling Outliers. Organizational Research Methods, 16(2), 270- 301.

Allison, P., 2000. Multiple imputation for missing data: A cautionary tale. Sociological Methods and Research, 28, 301–309.

Almansour, B. Y. (2015). Empirical model for predicting financial failure. American Journal of Economics, Finance and Management, 1(3), 113-124.

Amit, R., & Wernerfelt, B. (1990). Why do firms reduce business risk?. Academy of Management Journal, 33(3), 520-533.

Anderson, R. C., Mansi, S. A., & Reeb, D. M. (2004). Board characteristics, accounting report integrity, and the cost of debt. Journal of Accounting and Economics, 37(3), 315-342.

Ball, R., Gerakos, J., Linnainmaa, J.T. & Nikolaev, V. (2020). Earnings, retained earnings, and book-to-market in the cross section of expected returns. Journal of Financial Economics, 135(1), 231-254.

Bang, V. V. & Joshi, S. L. (2010). Market expansion strategy–performance relationship.

Journal of Strategic Marketing, 18(1), 57-75.

Barney, J. (1991). Firm resources and sustained competitive advantage. Journal of management, 17(1), 99-120.

Beal, M.J., Ghahramani, Z., & Rasmussen, C.E. (2002). The Infinite Hidden Markov Model.

Bendel, R. B., Higgins, S. S., Teberg, J. E. & Pyke, D. A. (1989). Comparison of skewness coefficient, coefficient of variation, and Gini coefficient as inequality measures within populations. Oecologia, 78(3), 394-400.

Bhattacharya, A., Misra, S. & Sardashti, H. (2019). International Journal of Research in Marketing. 36(4) 509-527.

Blagus, R., & Lusa, L. (2017). Gradient boosting for high-dimensional prediction of rare events. Computational Statistics and Data Analysis, 19-37.

Borenstein, M., Hedges, L.V., Higgins, J.P.T. & Rothstein, H.R. (2010). A basic introduction to fixed-effect and random-effects models for meta-analysis. Research synthesis Methods, 1, 97-111.

Table of Contents

Abstract

Acknowledgement