Loading the data
setwd("d:/my/documents/UvA/Advanced Corporate Finance/Assignment 2")
data1 <- read.delim("group_project_data.txt",stringsAsFactors = F)
data1$gvkey<-as.factor(data1$gvkey)
data1<-data1[!duplicated(data1[c("gvkey","year")]),]
#names(data1)
library(estout)
library(plm)
library(psych)
1. Desciption of what it means for a company to be financially constrained. Why might such firms save some of their cash holdings
- Companies that anticipate financial difficulties in the future could be considered ‘constrained’
- the pecking order theory restricts the ability of a company to raise new equity
- Companies that do not have credit rating or have a bad one could be considered as constrained, and could be expected to be more susceptible to cashflows and internal financing.
2. Create variable that represent financial constraint of lack of it using the data provided.
data1$rating_constraint<- as.numeric((!grepl(("BBB|A"),data1$credit_rating)))
dataConstraint <- data1[data1$rating_constraint ==1 ,] # creting separate datasets
dataNotConstraint <- data1[data1$rating_constraint ==0 ,]
- there are many approaches to define what is financially cnstrained company. I.e. an index could be constructed, companies could be selected on a basis of their scaled payout ratio, or the firm size could also be a proxy.
- Here(in this assignment), companies that do not have a credit rating or have one but is below investment grade(BBB) are considered ‘financially constrained’. The rest are ‘not constrained’
- Other criteria could also be computed
3. Create variables that measure corporate investment, investment opportunities and cashflows. The variables should be ‘scaled’, i.e. divided by total assets.
#creating ne variables and winsorizing them
data1$capex_scaled <- winsor(data1$capital_expenditures/data1$total_assets,na.rm=T,trim=.01)
data1$Qratio <- winsor((data1$market_value + data1$long_term_debt)/data1$total_assets,na.rm=T,trim=.01)
data1$cashflows_scaled <- winsor(data1$cashflows/data1$total_assets,na.rm=T,trim=.01)
data1$cash_scaled <- winsor(data1$cash/data1$total_assets,na.rm=T,trim=.01)
- The variables are created straight-forwardly, as can be seen from the code above. The investment and cashlows are scaled, i.e. divided by total assets.
- The Q ratio is computed by adding the long term debt to the market capitalization of a company and dividing that sum by book value of total assets. One possible problem is that we use the bookvalue of debt and not its market value.
- The variables are also ‘winsorized’
4. Running regressions of investment variable on investment opps and cashflows
model.1panel <- plm(capex_scaled ~ Qratio + cashflows_scaled, data=data1, model="within", effect = "time")
model.1cross_sec <- lm(capex_scaled ~ Qratio + cashflows_scaled, data=data1)
model.1panelCon <- plm(capex_scaled ~ Qratio + cashflows_scaled, data=dataConstraint, model="within", effect = "time")
model.1panelNotCon <- plm(capex_scaled ~ Qratio + cashflows_scaled, data=dataNotConstraint, model="within", effect = "time")
model.1panelModerate <- plm(capex_scaled ~ Qratio + cashflows_scaled, data=dataModerate, model="within", effect = "time")
| Qratio |
0.00255*** |
0.00248*** |
-0.00581 |
0.00303*** |
|
(0.00074) |
(0.00071) |
(0.0038) |
(8e-04) |
| cashflows_scaled |
0.06319*** |
0.06297*** |
0.1563*** |
0.06148*** |
|
(0.01161) |
(0.01123) |
(0.05219) |
(0.01277) |
| (Intercept) |
|
0.05289*** |
|
|
|
|
(0.00213) |
|
|
| R^2 |
0.01934 |
0.01859 |
0.02685 |
0.02044 |
| adj.R^2 |
0.01769 |
0.01791 |
0.02042 |
0.01836 |
| N |
2892 |
2892 |
472 |
2420 |
| Standard errors in parentheses |
|
|
|
|
- Panel data regression with fixed effect is estimated. And a normal OLS for a comparison as well. The panel data regression estimates the fixed effects, but also ‘time’ effects are added, since does no wrong and it is better to account for a time trend.
- R squared is low for all the regressions
- The effect of the cashflows variable is significant for the sample of companies, and is much larger than that of the Q ratio. There are several different explanations to that fact. The investement theory suggests that the Q ratio should determine the level of investment, but in reality most studies find that it has little effect. Our coefficient of 0.00255 is actually quite small, and is consistent with the findings of Peters and Taylor(2016) which found the effect Q of physical assets to be 0.006. On the other hand the theory suggests that numerous other things matter for the investment. In our case, naturally it could be that cashflows are also a factor in determining the level of investment. The coefficient is much higher than that of Q i.e. 0.06319 and it is significant. So we could conclude from this that both Q and cashflows matter for determining of the investment, but the effect of the latter is much higher.
- The effect of the cashflows is much larger for financially constrained firm. That is intuitively what we could expect, i.e. financially constrained firms should be more sensitive to available cash than the unconstraint. Our coefficients are highly signficant in both cases .15 and .06 respectively, suggesting a marked difference. The same result is obtaind in the paper of Fazzari, Hubbard, and Petersen (1988). We should also note that with a different division of the companies different results could be observed. lternative results are privided in the appendix.
5. Describe one econometric problem and how it can bias the obtained estimates
- Inability to measure correctly the Q ratio. This is a widely held view–we use proxies for the replacement value of firm’s assets. The book value of the assets is obviously an approximate measure. Furthermore the debt is also with its face value, which is not precise. Finally, the market value or the market cap could as well reflect the sentiment of the market as a whole.
- Perennially, the omitted variable bias could also play a role here. It is possible that some other factor also influence the level of investment.
6. Running regressions of cash holdings variable on investment opps and cashflows
model.2panel <- plm(cash_scaled ~ Qratio + cashflows_scaled, data=data1, model="within", effect ="time")
model.2cross_sec <- lm(cash_scaled ~ Qratio + cashflows_scaled, data=data1)
model.2panelCon <- plm(cash_scaled ~ Qratio + cashflows_scaled, data=dataConstraint, model="within", effect = "time")
model.2panelNotCon <- plm(cash_scaled ~ Qratio + cashflows_scaled, data=dataNotConstraint, model="within", effect = "time")
| Qratio |
0.02884*** |
0.02898*** |
0.0072* |
0.02832*** |
|
(0.00137) |
(0.00133) |
(0.0043) |
(0.0015) |
| cashflows_scaled |
0.00758 |
-0.00361 |
0.15614*** |
0.03853 |
|
(0.02145) |
(0.02111) |
(0.05905) |
(0.02391) |
| (Intercept) |
|
0.06092*** |
|
|
|
|
(0.00398) |
|
|
| R^2 |
0.1491 |
0.14527 |
0.08795 |
0.14704 |
| adj.R^2 |
0.13657 |
0.14468 |
0.0675 |
0.13231 |
| N |
2928 |
2928 |
473 |
2455 |
| Standard errors in parentheses |
|
|
|
|
- The same regressions are estimated as in 4.
- R squared is much higher
- The coefficient of Q is highly significant, whereas the one of cashflows is not statistically different than zero for the regression on the whole sample.
- The effect of Q ratio is higher for the cash variable than for the investment
- Overall the effect of cashflows on cash holdings of a firm is small
- For constrained firms the effect of cashflows on cash holdings is much bigger than that of unconstrained– .15 agains 0 or insinificant coefficient. Consistent with the findings of Almeida(2004), although the dependent variable in their regression is the change in cash holdings. Nevertheless our finings suggest that the cash holdings of constrained firms are more sensitive to cashfows than that of unconstrained. The numbers of Almeida(2004) for bond rating criterion are 0.05 and insignificat number for unconstrained companies.
Appendix. Using KZ index to divide companies
#creating a KZindex variable
data1$KZindex <- -1.002*data1$cashflows + .283*((data1$market_value + data1$long_term_debt)/data1$total_assets) + 3.139*data1$long_term_debt - 39.368*data1$dividends - 1.315*data1$cash
dataWithoutNA <- data1[!is.na(data1$KZindex),]
dataConstraint <- dataWithoutNA[(dataWithoutNA$KZindex< quantile(dataWithoutNA$KZindex,.3)),]
dataNotConstraint <- dataWithoutNA[(dataWithoutNA$KZindex> quantile(dataWithoutNA$KZindex,.7,na.rm=T)),]
dataModerate <- dataWithoutNA[((dataWithoutNA$KZindex< quantile(dataWithoutNA$KZindex,.7,na.rm=T)) & (dataWithoutNA$KZindex> quantile(dataWithoutNA$KZindex,.3,na.rm=T))),]
| Qratio |
0.00255*** |
0.00248*** |
-0.00101 |
0.00394*** |
-0.00501 |
|
(0.00074) |
(0.00071) |
(0.00124) |
(0.00107) |
(0.00342) |
| cashflows_scaled |
0.06319*** |
0.06297*** |
0.22505*** |
0.00784 |
0.30512*** |
|
(0.01161) |
(0.01123) |
(0.02227) |
(0.01709) |
(0.04699) |
| (Intercept) |
|
0.05289*** |
|
|
|
|
|
(0.00213) |
|
|
|
| R^2 |
0.01934 |
0.01859 |
0.14039 |
0.01444 |
0.068 |
| adj.R^2 |
0.01769 |
0.01791 |
0.11148 |
0.01203 |
0.05395 |
| N |
2892 |
2892 |
874 |
1137 |
852 |
| Standard errors in parentheses |
|
|
|
|
|
- Dividing the companies into three categories based on the so called KZIndex yields somewhat controversial results
- For moderately constrained companies cashflows actually does not matter for the investment, since the coefficient is insignificant
- Unconstrained firms, according to this regression have the highest and significant coefficient, followed by the constrained firms .30 and .22 respectively.
- The differencies of the results could be explain by the construction of the index. It is possible that it picks different effects, hence the results. With the credit ratings criterion, actually it appears clearer what is constrained/unconstrained firm.
- Neverthelass a different picture emerges