Introduction
The main purpose of this paper is to establish whether there is a link between financial conservatism and financial distress, namely whether financial distress leads to the adoption of financially conservative policies. This is a particularly poignant period to try to establish such a link as should we find one exists it would allow stakeholders (shareholders, employees, suppliers) to predict how firms would be governed in the aftermath of the worst financial crisis since 1929. Their interest in financial policy is stimulated by the significant economic and social costs that they would bear as the result of a company not surviving a period of financial distress and the ensuing bankruptcy costs. The investigation into a possible link between financial distress and financial conservatism is not new, many previous research papers have used a variety of methods (probit analysis, logit analysis, multiple discriminant analysis, univariate analysis) and tried to explain the existence of financially conservative policies, with reasons for their being ranging from differences in the ownership (Gammie 1986), to population characteristics (Jones 1994).
Theoretical Framework
Financial conservatism
One challenge is to correctly define the term financial conservatism, previous research has classified financially conservative firms as being those that persistently hold high cash reserves (Mikkelson & Partch 2003) or those that have historically low leverage levels (Minton and Wruck 2001). The motivation for holding large cash reserves has been attributed to a firm's desire to
provide a low cost of financing to protect shareholders interest (Myers and Majluf 1984, Almeida 2004);
serve the managers interest by consuming private benefits easily (Jensen 1986);
serve as a precaution for potential future problems (Calmès 2004);
reduce tax costs associated with repatriation of foreign income (Foley et al 2006).
Likewise many reasons exist for the desire of firms to keep leverage low. Some of the most popular reasons why they hold this have been attributed to
providing shareholders with flexibility in financing future investments (Myers 1984);
protect managers against the expected costs of financial distress (Berger et al 1997).
According to Iona et al (2004) some authors claim that cash-conservatism and leverage-conservatism may act as substitutes (Kaplan and Zingales, 1997; and Kashyap, Lamont and Stein, 1994), therefore there is no point in investigating them separately. The other argument of Iona et al (2004) that we find very reasonable is that substantial cash balances may be kept to avoid debt financing. For these reasons instead of looking at cash and leverage conservative companies separately we examine financially conservative companies, which are as defined by Iona et al (2004) the companies, which are at the same time cash and leverage conservative.
Iona et al (2004) in their paper used discriminant analysis based on the non-parametric estimate of the distribution of cash holdings and leverage when determining financially conservative companies, but the methodology choice in comparison to fixed classification rule, didn't influence the conclusions made despite its theoretical advantages. We have therefore decided to use a simpler fixed classification rule, according to which we consider the company to be cash-conservative if it holds more than 25% of its assets in cash and cash equivalents and the company is considered to be leverage-conservative if its annual ratio of total debt to total assets is low enough to be among the first 20% of all companies.
To capture persistency Iona et al (2004) split their data into non-overlapping 3-year panels, where each panel is treated as an observation and company is considered persistently cash or leverage conservative if it stays in the same panel for all 3 years. We decided to deviate from this methodology of capturing persistency, because, to our opinion, in this case valuable observations are lost. If the company was conservative for the first 2 years inside the panel these years are not considered conservative even if the previous panel included 3 conservative years. This means that the last 2 years of 5 years long conservative period will be lost. We consider the company to be conservative on the third and following consecutive years to keep these observations. We exclude the first 2 years because we believe that it may take time for the financial indicators to change when the company adopts financial, cash or leverage conservative policy and we believe that this is a more efficient way of distinguishing between transitory and permanent conservative policies. As was mentioned before, the company is considered financially conservative in the particular year if it is cash and leverage conservative at the same time at the given year.
Financial distress
Financial distress is a condition of the company when it faces difficulties paying off its financial obligations, it has been defined and determined in many past papers and taken into account many variables. The extreme case of financial distress is bankruptcy, but the more common scenario is when company experienced a write down of their debt or a forced sale.
Size and age of the firm are suggested to determine the likelihood of a company entering financial distress, there is a suggestion that younger and smaller firms are likelier to become financially distressed as they lack the experience and have limited information (Audretsch and Mahmood, 1995; Honjo, 2000). Furthermore, Dyrberg (2004) suggested two more hypothesis related to firm size and the probability of failure and financial distress, firstly that the effect of the firms size on the likelihood of financial distress is U shaped as small firms are less resistant to financial shocks whilst large firms are likelier to enter financial distress because of diseconomies of scale such as poor intra firm communication between management and staff and general inflexibility in the firm. Dyrberg's (2004) second hypothesis was that the probability of financial distress is a decreasing function of firm size. We decided not use size characteristic of the company in our regression due to its ambiguous and possibly non-linear relationship with financial distress.
McCrae, Davy et all (2007) suggested that financially distressed (defined as bankrupt in their study) companies show certain characteristics such as lower profitability, higher leverage, lower past excess returns and larger size compared to active companies. Other measures of financial distress include the fixed to variable costs ratio, debt management ratios, asset utilization ratios amongst others.
We decided to use bankruptcy prediction models to indicate companies being in financial distress. Models like Altman's, Springate's or Fulmer's model, clearly define the relationship of some ratios to the likelihood of the company to go bankrupt, which is the extreme case of financial distress. These are classic models that have been tested many times and gave good results, therefore we are quite confident that the relationship of variables in these models with financial distress is linear, unlike the company's size. Unfortunately, we did not have all the data required by any of these models, we therefore decided to search for common financial indicators used in these models and select those which are given the highest weights as they should be the main indicators of financial health of the company.
Altman's model
Z = 1.2A + 1.4B + 3.3C + 0.6D + 0.999E , where
A = Working Capital/Total Assets
B = Retained Earnings/Total Assets
C = EBIT/Total Assets
D = Market Value of Equity/Book Value of Total Debt
E = Sales/Total Assets
The larger is Z the better is the company's condition; 2.674 is considered to be a threshold point, below which the company is considered failed.
Springate's model
Z = 1.03A + 3.07B + 0.66C + 0.4D , where
A = Working Capital/Total Assets
B = EBIT/Total Assets
C = EBT/Current Liabilities
D = Sales/Total Assets
The larger is Z the better is the company's condition; 0.862 is considered to be a threshold point, below which the company is considered failed.
Fulmer's model
H = 5.528 (V1) + 0.212 (V2) + 0.073 (V3) + 1.270 (V4) - 0.120 (V5) + 2.335 (V6) + 0.575 (V7) + 1.083 (V8) + 0.894 (V9) - 6.075 ,where
V1 = Retained Earning/Total Assets
V2 = Sales/Total Assets
V3 = EBT/Equity
V4 = Cash Flow/Total Debt
V5 = Debt/Total Assets
V6 = Current Liabilities/Total Assets
V7 = Log Tangible Total Assets
V8 = Working Capital/Total Debt
V9 = Log EBIT/Interest
The larger H is the better the condition in which the company finds itself. A value 0 is considered to be a threshold point, below which the company is considered bankrupt and failed.
We have established that the following elements are most often used in the stated models:
Working Capital, given high weights by all 3 models;
EBIT, which is given highest value by Altman and Springate;
Sales, which are usually given quite low weight, but is used in all 3 models;
Retained earnings, highly valued by Altman and Fulmer.
In our dataset we have the following ratios that account for some of the above financial indicators used in bankruptcy prediction models, although sometimes do not exactly duplicate them:
The ratio of current assets minus current liabilities and total cash to total assets (liq2)
The ratio of earnings before interest payments and tax (EBIT) to total assets (profit2)
The ratio of total sales in constant prices to total assets in constant prices (sales) = exp[The logarithm of total sales in constant prices (logsal) - the logarithm of total assets in constant prices (logass)]
Hypothesis
In our paper we test the hypothesis that financially conservative companies adopt this policy because they are suffering from financial distress.
Null hypothesis: financially conservative companies adopt this policy NOT because they are suffering from financial distress.
Alternative hypothesis: financially conservative companies adopt this policy because they are suffering from financial distress.
To test the null hypothesis we regress the following equation:
fcons2 = β0 + β1 x liq2 + β2 x profit2 + β3 x sales + ɛ, where
fcons2 is a dummy variable, which is 1 if the company in this particular year is financially conservative and 0 if it is not. To test more specific cases of financial conservatism we may substitute the dependent variable with ccons2, which stands for cash conservatism, or lcons, which stands for leverage conservatism.
We expect all 3 independent variables to be negatively related to financial distress as they are given positive weights in all 3 bankruptcy prediction models. If the relationship of these variables is the same for financially conservative companies it will mean that there is positive relationship between financial distress and financial conservatism.
Intuitively this makes sense, the higher the liquidity of the company, the quicker it can translate assets to cash and meet the challenges of financial distress and avoid bankruptcy. Likewise, higher sales and profitability would suggest that the company is performing well and moving away from bankruptcy, some of its profits would be retained and reinvested to generate sales growth. Furthermore, if the company was struggling with its balance sheet but making profits, it would find external financing easier to obtain and avoid any bankruptcy challenges.
Empirical Framework
Huselid and Day (1991) claim that using OLS to estimate a binary dependent variable violates three important assumptions:
"1. Predicted values can fall outside the logical 0-1 boundaries, yielding meaningless results (Amemiya, 1981, pp. 1, 486; Maddala, 1983, p. 16).
2. Heteroscedasticity and nonormality of the errors invalidate the coefficient t-tests for the independent variables (Doran, 1989, p. 315; Maddala, 1988, p. 269).
3. Estimates of the marginal effects of an independent variable are biased because they depend on the mean value of the dependent variable (Doran, 1989, p. 316; Maddala, 1983, p. 24)."
We decided to use the probit regression to avoid these drawbacks of OLS, although results of logistic regression are not so straightforward to interpret. We need only to find out if there is relationship between dependant in independent variables and if it is positive or negative, this information in easily available from probit regression output.
Data
Our dataset is an unbalanced panel data of 1,195 non-financial UK companies from year 1984 to year 2002 totalling to 14638 observations. We balanced the data between starts and ends adding 8 empty observations. These companies are from 15 industries and 157 sectors.
As is clearly seen from the below histogram our observations are well spread across different industries. Below we provide description of the variables that we used in our hypothesis testing.
The median value of the ratio of holdings of cash and cash equivalents to total assets is far less than the mean value in our data which suggests that a lot of values are below the mean and suggests we have relatively few high values. The skewness value of 2.32 suggests that this is the case. The kurtosis value suggests that the variance is down to fairly infrequent and extreme deviations as opposed to modest size variations. The Jarque-Bera test result is very high and suggests that the data is departed from normality.
The mean of the ratio of total debt to total assets is greater than the median for the data of this variable, which suggests that there will be a positive tail on the frequency distribution graph. The skewness value confirms this, although it is higher than anticipated because of a large maximum value relative to the mean and the kurtosis value shows that the variance is down to infrequent extreme deviations. The standard deviation is small for the data set. The Jarque Bera test shows that the data is departed from normality, but not by much.
The data for the ratio of earnings before interest payments and tax to total assets indicates that there will be a large negative skewness due to a larger number of low values compared to high. The mean is less than the median and this supports the value of the skewness. There is also a very large spread between the mean, median and minimum value which will indicate that there are extreme outliers which will account for a large part of the variance in the data which is confirmed by an extremely high kurtosis value. The standard deviation is also high indicating a greater dispersion in the data. The Jarque-Bera test is extremely high as expected due to the extremely low minimum value and the data is very far from normality.
The median value of the ratio of total sales in constant prices to total assets in constant prices is smaller than the mean value meaning that there will be a positive skewness as the tail of the frequency distribution will be positive due to relatively few low values which is confirmed by the value of the skewness. The skewness and kurtosis are extremely high because of the spread between the minimum and maximum value and that a lot of the variance is due infrequent and extreme values. The Jarque-Bera test value shows that the data is far from normality and the spread of the data is large as indicated by the standard deviation.
The mean of the ratio of current assets minus current liabilities and total cash to total assets is smaller than the median for this data suggesting relatively few lower values and we would expect a negative tail and skew as a result which is confirmed by the skewness statistic. The standard devaition shows that the dispersion of the data isn't too great. The kurtosis is not extremely high but none-the-less indicates that there are some infrequent extreme values which can explain the variance. The Jarque-Bera test shows the data is very far from normality.
Results
By regressing the equation described above with financial conservatism as dependent dummy variable (fcons2) gives the following output:
Variable
Coefficient
Std. Error
z-Statistic
Prob.
C
-1.634151
0.043782
-37.32443
0.0000
LIQ2
-0.085881
0.025812
-3.327127
0.0009
PROFIT2
0.654608
0.151011
4.334828
0.0000
SALES
-0.276318
0.033575
-8.229811
0.0000
McFadden R-squared
0.025760
Mean dependent var
0.029148
S.D. dependent var
0.168227
S.E. of regression
0.167773
Akaike info criterion
0.257296
Sum squared resid
411.2666
Schwarz criterion
0.259373
Log likelihood
-1876.188
Hannan-Quinn criter.
0.257986
Restr. log likelihood
-1925.795
LR statistic
99.21515
Avg. log likelihood
-0.128374
Prob(LR statistic)
0.000000
Obs with Dep=0
14189
Total obs
14615
Obs with Dep=1
426
All variables are significant at 1% significance level. Low McFadden R-squared doesn't mean that we are doing something wrong, because we are not trying to explain the whole variation of fcons2, we just need to find correlation of variables with it. Such a good significance level for all betas allows us to rely on these results. Wald test also shoes that at 1% significance level all betas are not equal to zero and therefore influence the dependent variable.
Profit2 is the only variable, whose beta doesn't allow us to reject the null hypothesis, but we have an explanation of such result. The more the company earns relative to its size, the higher amount of cash it receives and the higher should be the proportion of cash among total assets. The company may use this cash in two ways: keep it (cash conservatism) or use it to finance some projects, avoiding debt financing (leverage conservatism). Both these ways of using earnings increase the probability of the company to be classified as financially conservative, therefore profit2 influences both financial conservatism and financial distress in opposite directions and is an inappropriate variable to use for our regression. Sales variable's beta is also influenced by these factors, therefore we do not have enough reliable information in the output to reject the null hypothesis.
Regressing with cash conservatism as a dummy variable gives the following output:
Variable
Coefficient
Std. Error
z-Statistic
Prob.
C
-1.368011
0.029973
-45.64150
0.0000
LIQ2
-0.099497
0.020809
-4.781432
0.0000
PROFIT2
-0.018481
0.076339
-0.242095
0.8087
SALES
-0.189234
0.019748
-9.582520
0.0000
McFadden R-squared
0.019633
Mean dependent var
0.053575
S.D. dependent var
0.225185
S.E. of regression
0.224005
Akaike info criterion
0.410165
Sum squared resid
733.1557
Schwarz criterion
0.412242
Log likelihood
-2993.280
Hannan-Quinn criter.
0.410855
Restr. log likelihood
-3053.224
LR statistic
119.8878
Avg. log likelihood
-0.204809
Prob(LR statistic)
0.000000
Obs with Dep=0
13832
Total obs
14615
Obs with Dep=1
783
Only two variables: liq2 and sales have significant betas, both of which contradict with the null hypothesis, but we still decided not to reject the null hypothesis for cash-conservative definition of financial conservatism as well because all variables should have significant and negative betas and as discussed above sales and profit2 variables influence the company being in distress or not and it's likelihood to be financially conservative in opposite directions.
For leverage conservatism result is the same as for financial conservatism with one exception:
Variable
Coefficient
Std. Error
z-Statistic
Prob.
C
-1.342525
0.030358
-44.22287
0.0000
LIQ2
0.171228
0.072877
2.349542
0.0188
PROFIT2
0.825945
0.106773
7.735512
0.0000
SALES
-0.055645
0.018348
-3.032719
0.0024
McFadden R-squared
0.008772
Mean dependent var
0.096066
S.D. dependent var
0.294691
S.E. of regression
0.293676
Akaike info criterion
0.627700
Sum squared resid
1260.133
Schwarz criterion
0.629778
Log likelihood
-4582.920
Hannan-Quinn criter.
0.628390
Restr. log likelihood
-4623.475
LR statistic
81.11165
Avg. log likelihood
-0.313576
Prob(LR statistic)
0.000000
Obs with Dep=0
13211
Total obs
14615
Obs with Dep=1
1404
The only difference is that now the liq2 variable has a positive beta, but is insignificant at 1% significance level that we use throughout this paper due to having a very large sample.
Improving the model
There is a number of variables that we believe affect if a company is in distress or not which we could not include into our model as we didn't have sufficient data. The example of such variables is remaining variables from bankruptcy prediction models.
Size and age of the firm are suggested to determine the likelihood of a company entering financial distress with Audretsch and Mahmood (1995), Honjo (2000) and Dyrberg (2004) all suggesting smaller firms have a higher probability of becoming financially distressed and Dyrberg (2004) also suggesting that very large firms face financial distress due to diseconomies of scale. We might remove the largest companies from the sample and add amount of assets into the equation to avoid letting the hypothesis of Dyrberg (2004) about the U-shaped relationship between financial distress and size of the company adversely affect the regression results.
The state of the economy/economies in which the firm operates and generates sales from should also be taken into account, we feel it would be appropriate to remove observations highly influenced by the adverse economic conditions from the sample. This may be done by using only years with positive GDP growth rate in UK, although we some of the companies in the dataset generate high proportion of sales abroad.
The other absolutely different way of testing the null hypothesis is to indicate those companies, which had forced sales, went bankrupt in the following year or wrote down some debt. Such companies may be classified as being in financial distress and the probability of being at the same time financially conservative and in financial distress may be calculated.
Conclusions
We couldn't reject the null hypothesis because of insufficient data in our dataset. We have ideas about improvement of our model and making the hypothesis testing more reliable. Regression results indicated that profitability has positive correlation with financial conservatism although the correlation with financial distress should be negative. We have a hypothesis that explains such behavior, but according to this hypothesis sales variable should also influence both financial conservatism and financial distress in opposite directions and therefore cannot be used in regression. Liquidity ratio can't alone indicate the relationship between financial conservatism and financial distress, therefore we decided not to reject the null hypothesis about the lack of such relationship. Further investigation in this field should be done with a different dataset, which should include additional variables.
