5. Results
5.2 Switching costs estimated from firm’s bank connections
5.4.2 Results of the estimation
Table 4: Estimation of firms’ switching costs dependency on firms’ characteristics.
Table 4 shows the results of the regression. The first column shows the name of the variable, the third and fourth show the actual parameter estimate, of which only the sign is of interest. The last two columns show the t value and corresponding P-‐value. Below will individual interpretations of each parameter estimate be presented.
The intercept of the estimation is not of interest, both because the level of switching costs is not interesting in this estimation, but also because it does not make economic sense to consider the switching costs of a firm where independent variables are zero. Generally most of the variables appear to be significant and the standard errors are quite low, which is most likely because of the large dataset used for the estimation.
The parameter estimates of the balance and capital are both positive and very significant compared to the other variables. Firms with a larger balance or capital are thus predicted to have larger switching costs. So the size of the firms and their switching costs are, according to the estimation, positively correlated. It seems reasonable that both parameters have the same sign, as both variables are expressions of size of the consumers. The positive relationship is also what would be expected from theory. The parameter of capital is larger and more significant than balance. This could be because capital is a better measure for firm size than the balance. The balance includes all the firm’s operations and can easily be affected by both the firm’s accounting and financial policies. The capital is more narrowly defined, and therefore better at predicting the level of switching costs of the firm.
The reason for this positive correlation can be found on both the supply and demand side. Larger firms are likely to require more banking services and are therefore relatively more important for the banks. Therefore banks put more effort into keeping the larger customers. The high volume of
business probably also makes it more inconvenient to switch, but may also increase the firms incentive to choose the best current contract. The last effect seems to be offset by the others. All in all, these effects and maybe other effects, reduce the likelihood of firms switching bank.
The variable profit is a dummy variable that is one if a firm made a profit in their latest financial report. The parameter is negative and significant. According to the estimation, profitable firms have lower switching costs than firms that did not make a profit. This result is in line with the theoretical prediction. Unprofitable firms are generally not desirable customers in banks, as they have a higher risk of defaulting on their loans. The most obvious reason for the relationship between switching costs and firms’ profitability is that the firms and the banks have obligations to each other through a contract, and that there is very low supply of credit for unprofitable firms. Competing banks are not likely to offer loans to unprofitable firms, let alone attractive contracts, so the firms often have to stay with their current bank connection. The current bank often have an interest in continuing the relationship with the firm if it is unprofitable, to lower the risk of the firm defaulting on its loans. If the bank stop supplying credit to the firm, and the firm is unable to get financing elsewhere, the firm may default and be unable to pay back its debt in full.
The debt of firms is correlated with the other variables through the size of the firm, but it is also the product being traded between the firm and the bank. This makes it hard to predict the variables’
relationship with the firms’ switching costs. The estimation reveals that the parameter is negative and very significant. Thus the more debt a firm has, the lower is its switching costs.
There are at least two effects that contribute to the parameter estimation. The first is the size effect of the firm. The larger a firm is, the larger is their debt generally. Other variables, such as balance and capital that are also linked to the size of the firm, revealed a positive correlation between size of the firm and switching costs of firm. The other effect is a consequence of the increased size of the contract between the bank and the firm. The larger a potential contract is, the more incentive does the competing banks have to offer a good contract to the firm, which can make firms’ incentive to switch larger. The firms also have a larger monetary incentive to get the best possible contract, due to the large volume. The current bank connection of the firm obviously has the same incentive to offer the firm a good contract, but the increased competition between banks and the strong incentive for the firm to choose the best current contract seems to exceed this effect. The size effect, that definitely is present in the estimation, is offset either by the before mentioned effect or by other unknown effects.
The equity of the firm is estimated to be positively correlated with the switching costs of the firm.
The larger a firm’s equity is, the larger is the estimated switching costs of the firm. Equity is also positively correlated with the size of the firm, which is positively correlated with switching costs. But equity is also an indication of firms’ ability to absorb losses. A high equity is therefore also a sign of a healthy firm, which is negatively correlated with switching costs as with the profit variable.
According to the sign of the parameter, the size effect is the dominating factor. This is likely because the variable does not compare the equity level to other factors, so it is not possible to infer if the equity for a given firm is high or low. The solvency ratio does exactly that. It relates equity to the assets and therefore excludes the effect of the size of the firm from the variable.
The solvency ratio is a continuous logarithmically transformed variable. The estimated parameter is negative and significant. It is one of the least significant variables with a standardized P-‐value around 0.7%, which is still well below the conventional cutoff value of 5%. According to the estimation the higher a firm’s solvency ratio is, the lower is the firm’s switching costs. The solvency ratio is an expression for a firm’s ability to meet its obligations. It is related to the profitability measure, as both variables measure the firms’ health. It is therefore not surprising that both have the same sign.
Generally the estimation reveals that healthier firms have lower switching costs, most likely because they are more attractive customers to competing banks. The reason for the low significance of the variable is most likely due to high variation in solvency ratios of otherwise similar firms. This variation is caused by sector and industry differences, as well as the nature of the ratio. If a firm has few assets, which is the denominator of the fraction, the equity has a high impact on the total ratio, and vice versa.
The number of bank connections that each firm has in the dataset is not a financial figure like the other variables. It is mainly included to reduce the effect of some firms weighting more than others in the estimation, if they have more than one bank connection. The parameter is negative, but not very significant. The adjusted P-‐value is slightly above 10%, so the variable would usually be
considered insignificant. The standard error does however reveal that the parameter estimate with a high probability is negative, which is an interesting result. The number of bank connections a firm has is possibly correlated to the size of the firm, but it is not clear how strong this effect is. Most firms can get their demand fulfilled by one bank, but larger firms may want more bank connections, to spread their risk and dependency across several banks. The logical relationship between the number of bank connections that a firm has, and the switching costs of the firm, is negative. Multiple bank connections indicates that the firm is not loyal to one bank and therefore have lower costs of switching. If a firm has more than one bank connection, then it will be easier to close a bank connection with a bank, which in the dataset will be interpreted as a switch. This is not a desired feature, but it is the disadvantage of including firms with more than one bank connection. The last mentioned effect dominates the size effect in the dataset, according to the estimation.
The last parameter estimates are those of the dummy variables containing information about the year of establishment. The default category is the group consisting of firms established before 1950.
All parameter estimates are therefore in relation to this group. All the parameter estimates related to the year of establishment are negative, meaning that the other groups of firms has lower switching costs than the default group. It is expected that, if there is a significant relationship, then it is a positive correlation between the number of years a firm has existed and the switching costs of the firm. The parameter estimations are thus in line with the expectations. The group of firms
established between 1950 and 1959 has a P-‐value of 0.23%, which yields a standardized P-‐value
slightly above 5%. The group of firms established between 1960 and 1969 also has a standardized P-‐
value above 5%, which suggest that the parameter estimates are not significant, and consequently that the switching costs of the firms established between 1950 and 1969 are not significantly
different from the switching costs of firms established before 1950. The three groups are arbitrarily composed and they all contain firms that are very mature, it is therefore reasonable to accept that they do not have different switching costs. The other groups have parameter estimates that decrease as the ages of the firms’ decreases. According to the estimation, the more recently a firm was
established, the lower switching costs does it have, which is also in line with the expectations of the relationship. The relationship can be partly explained by the likely correlation between the amount of time a firm has been a customer at a bank and the year a firm were established, caused by the fact that firms generally do not switch banks very often. The longer a firm has been a customer at a bank, the higher switching costs are the firm expected to have.
The variables of the estimation are generally all relatively highly correlated, as firms’ characteristics are all related to size of the firms, which may lead to multicollinearity. The issue of multicollinearity would be an important issue if a smaller sample size were used for the estimation. Multicollinearity increases the standard errors, because it is not clear which of the independent variables that are responsible for the variation in the dependent variable. This effect is mitigated by the large sample size, as can be seen from the standard errors of the estimation.
The coefficient of determination, 𝑅!, is slightly above 2%. It is very low, which means that the model does not do a very good job at predicting the observed values. Thus the characteristics of firms, used in the estimation, are not good predictors of the firms’ switching costs. It was not expected that the firm characteristics could explain very much of the variation in the consumer switching cost, so a low coefficient of the determination was expected. The dataset include all firms, without controlling for outliers, other than transforming the variables. Looking at the data, it is evident that there are many large outliers, as well as significant variation across industries. This may also contribute to a low the coefficient of the determination. The focus of the estimation was not to set up a model to predict the consumer switching costs on the basis on consumers’ characteristics, but to examine if there were some relationship between the characteristics and the switching costs. If the model was constructed to predict the switching costs of consumers, the coefficient of determination could most likely be
increased if outliers were removed from the dataset, and the industry of the firms were controlled for.
Generally the parameter estimates are all very reasonable, and are consistent with both theory and expectations. The most important result of the estimation is that almost all of the variables are significant. While the variables does not explain very much of the variation in the switching costs, it is still notable that they have an influence on the switching costs of the firm.