1 Introduction

T

The meaning of the Fringe Benefits Tax (FBT), the tax payable, tax base and rates have been discussed in the article published in this journal on 16 August 2008. That article also discusses the research design and data collection of the larger study.

2 Research Approach and Collection Pattern

We discuss below the approach in this paper and the col lection pattern.

2.1 Collection Summary

FBT collection was Rs 4,772.3; Rs 5,323 and Rs 6,743 crore in the first three years of its operations, the financial years 2005-06, 2006-07 and 2007-08, respectively. As a percentage of total direct tax collection, it translates only to around 2.8%, 2.3%, and 2.2%, respectively in the three years. But, for a resource starved country, an additional Rs 6,000-7,000 crore is not a small sum. Further, FBT collections show an ABC pattern, similar to that of corporate income taxes. The collection data was a nalysed by classifying business/economic activities into 22 sectors of the economy. Banking, petrochemical, infotech, and insurance are found to be some of the important contributing sectors. Similarly, “Employee Welfare”, “Conveyance”, “Telephone”, “Running of Car”, “Sales Promotion” are some of the important contributing heads. It has been found that overall, the top 10 “heads” are contributing more than 90% and that the b ottom five “heads” are contributing around 2% of FBT collection. A “head-wise” and “sector-wise” summary has been incorporated in Tables 1 and 2 (p 61). A detailed analysis of sectorwise and head-wise FBT collection has already been done in a s eparate article (Kishore 2008).

Economic & Political Weekly

EPW

january 3, 2009

3 Statistical Tests and Inferences

There are certain questions and issues which have arisen during the preliminary analysis of the collection data. Some of these are: Are the overall collection patterns of FBT in the two years of its operation similar? Are these patterns similar even for different sectors of the economy? Can we conclude something about the collection pattern with certainty using some statistical test? Is the collection of FBT from an individual taxpayer dependent on the “economy sector” or on “heads” or on both? What do the top 10 and bottom five “heads” signify? Can the bottom five heads be removed from the FBT system without affecting the collection? Can we infer something about booking of expenses and uniformity of sample data? To address most of these issues, certain statistical tests have been conducted to get some rigorous and dependable inferences.

Statistical tests were conducted at three levels of collection data of the FBT. First, at the level of overall FBT collection to check equality of collection pattern under each head for the first two years. Second, tests were conducted separately for each of the ‘economy sectors’ to test the equality of proportion of collection under each head for the first two years. To measure the interaction between the “economy sectors” and “FBT heads”, a two- factor ANOVA was also conducted. Lastly, tests were conducted to see the homogeneity of proportion of collection in sample data for each combination of “head” and “sector”. For all the tests, collection of FBT from each “head” was converted into proportion of total FBT collection. Some other modifications were also made in the data which are explained in the coming paragraphs.

4 Test of Equality of Proportion Over the Years

This section presents the results of test of equality over the two years, 2005-06 and 2006-07.

4.2 Test Results and Conclusions

This test was conducted for each heads of FBT separately, thus totalling 17 tests. The results of these tests are shown in Table 1. (The table also shows percentage contribution by each head of FBT in the total collection for the financial year 2006-07 and 2005-06, although with some adjustments.) It can be seen that the null hypothesis is accepted in case of all the heads of FBT. This means that in totality, the proportion of contribution by each head of FBT in the total FBT collection is not significantly different for both the years. Further, the value of Z statistics is always less then

1.00 and in many cases it is even less than 0.50. This means that the null hypothesis would be accepted even at a stronger level of confidence. Therefore, it can be said that the proportion of collection by different heads in the two years has remained the same and statistical evidence to support this hypothesis is very strong.

Economic Political Weekly

EPW

january 3, 2009

This means that we have strong statistical evidence to conclude that the proportion of collection of FBT from different heads has remained same over the two years. This points towards an overall stability in FBT regime and FBT collection from the very first year of its operation.

5 Chi-Square Tests

This section carries out a test of equality of proportion over the years for each economy sector.

6 Test of Homogeneity of Sample Data

This section presents the analysis of variance (Anova) and chi-square test for the e conomy sectors and FBT heads.

TIFAC

6.1 Chi-Square Test for Each Combination

The first two levels of tests have shown broad homogeneity of collection data. The last levels of tests have been conducted to check the homogeneity of collection proportion in case of individual samples in a particular combination of sector and head. All 22 economy sectors are taken one by one and chi-square test for homogeneity of sample data has been conducted for each of its 19 FBT heads. It has been checked whether individual FBT p ayers in a particular combination of the economy sector and FBT head have a statistically similar pattern of contribution. For example, in the automobile-ancillary sector, wherein the sample consisted of data from 17 individual FBT payers, it is checked whether per cent contribution3 from a FBT head, say employee welfare from each of these 17 taxpayers in sample, is statistically homogeneous or different. In essence, it is tested whether the FBT collection proportion shown by individual samples in a particular combination of sector and head are statistically equal to the average p roportion of that particular combination. Since same types of business generally have similar kind of expense patterns, in the ideal situation, the sample data are expected to show some s tatistical homogeneity.

6.2 Two-Factor ANOVA

However, before doing these chi-square tests, a two factor ANOVA has been conducted to test whether there are differences in proportion of FBT collection from different heads and from different economy sectors – being two factors. The main idea is to test whether there are differences in proportion of FBT collection from different heads and from different economy sectors. In effect, it has been tested as to whether collection of FBT from a taxpayer is dependent on “economy sectors” or on FBT heads or on both and it what fashion. For this purpose, a two-factor ANOVA

Economic Political Weekly

EPW

with different observation per cell has been conducted. The combined effect of both these factors, beyond what is expected from the consideration of each effect separately, called the interaction effect, has been found by calculating mean squares of two factors and mean squares error.

The F-ratio for interaction effect has been calculated as 10.002. The critical value of F-ratio has been found to be 1.22 at the 0.05 level of significance, with the given level of numerator and denominator degrees of freedom.4 Thus, test statistics of F ration of 10.002 indicated a very strong interaction effect among the two factors – sector and head, leading to separate chi-square tests for all possible combinations of these two factors.

6.3 Classification of Expenses

The chi-square test is relevant in analysing another significant issue being classification and booking of expense. As of now, there is no standard procedure or classification system for booking of expenses and its accounting treatment by business organisations. It depends solely on the accounts/finance department to devise/determine a head, and book an expense under a head. No guidelines or accounting standard have been issued by Institute of Chartered Accountants of India (ICAI) for this purpose. What is generally found in the books of account of large organisations is a broad four or fivefold classification of all expenses into manufacturing, selling, employees, administrative and miscellaneous expenses heads. These broad heads are then s ubdivided into v arious specific heads for booking of expense and there is no uniformity even for naming of such a specific head. This gives complete discretion to an organisation for classifying and booking an expense. The classification issue is a complex one. It should also be realised that due to complex nature of modern business enterprises and diversity in nature and type of expenses incurred in the course of business, it is

Table 4: Petrochemical Sector – Summarised Results of Chi-Square Tests for Homogeneity of Sample Data

very difficult to have a practicable classification system and standardised b ooking of expense.

It has been noted during preliminary data analysis that out of the bottom five heads, four heads are those heads where the base/valuation rate is 50% of total expense. Further, these four heads are such where the expense made for these purposes can also be booked under other heads of expense, notably under employee welfare. The head employee welfare is a general type of head which can include expenses incurred for providing scholarship, festival celebration, etc, and doing so would perfectly be within the four corners of law. However, when done so, it will have the impact of reducing the FBT liability due to d ifferential valuation base for these heads of expense. Prima facie, it may be the reason why the heads with 50% base are least contributing and also why the head employee welfare is the largest contributing head. However, it would have been too naïve to arrive at such a conclusion on the basis of above simplistic notion.

The present tests could give us the required insight. It is safe to assume that there would be a kind of homogeneity in the nature of expense incurred by entities engaged in same e conomic/ b usiness activity. That is, for infotech-ITES sector as a whole, it can be assumed that expenditure incurred on some head, say tele phones as a proportion of total expense or some other similar parameter would be similar for many/most of individual taxpaying entities. Following this logic, if this being the case, collection of FBT from a head as a proportion of total FBT collection in case of each individual taxpayer in the sample should show statistical homogeneity/equality for each com bination of head and sector. If this is not the case, there is some indication to believe that the sample data are h eterogeneous and that perhaps booking of expense is arbitrary.

The chi-square statistics calculated are compared with the critical value of chi-square distribution for the required degree of freedom and confidence level (taken to be 95% in this case). For each of the economy sectors, sample sizes are different which gave different n and different values for degree of freedom as (n-1). The sample size of different sectors of the economy varies between 38 (banking) and 7 (steel). Accordingly, the critical v alues of X2 are different for different sectors of economy as they depend on size of sample also. Finally, if the test statistics, i e, X2 value calculated is less than the critical value, null hypothesis is accepted, otherwise it is rejected.

6.5 Test Results and Findings

A total of 418 chi-square tests were to be conducted, one each for each possible combination of sector and head (22 sectors × 19 heads). However, in 18 instances involving the heads “Free ticket” and “Rep, Runn, Dep of Aircraft” for different economy sectors, there were no data points and therefore, no test could be conducted. Thus total number of chi-square tests actually conducted are (400 = (22 × 19) – 18).

Test results for two sectors of the economy banking (18 tests) and petrochemical (19 tests), being the two largest contributing sectors, for all the heads are given in Tables 3 and 4 (p 63). In case of banking, null hypothesis has been rejected in all the 18 tests, i e, for all the heads implying thereby that the s ample data is not homogeneous even for a single combination. For the petrochemical sector, the null hypothesis has been rejected in 13 tests, i e, for 13 heads and accepted in six tests, i e, for six heads showing some homogeneity in sample data for accepted heads.

6.6 Summary of Results and Inferences

Thus, in total, 400 chi-square tests have been conducted for all combinations of sectors and heads. However, it is difficult to find meaningful patterns, inferences and insights form the results of such a large number of tests at first sight. Therefore, we look deeper into the results by summarising them. Table 5 (p 64) presents the results of all the 400 chi-square tests in the form of a matrix. The heads of FBT has been listed horizontally while the sectors of the economy have been shown in vertical fashion. Thus, each cell of the matrix represents the result of chi-square test for the sample data represented by that cell. R represents the cases when the null hypotheses have been rejected (and therefore, we concluded that the sample data is heterogeneous) whereas A represents the cases where the null hypotheses have been accepted (and we concluded that the sample data is showing homo geneity). X marks the cells where no test has been conducted due to lack of data. Out of total tests, in 99 instances, which is around 25% of total number of tests, the null hypothesis has been accepted. Therefore, overall, it can be concluded that there is not much statistical evidence to accept the null hypothesis and accordingly it is difficult to conclude that the sample data is homogeneous.

It is not the individual test results that are significant but the summary of these and the pattern of these results which throws valuable insights. From Table 5, we notice that the upper left corner has very few acceptances of null hypothesis whereas the occurrences of acceptances of null hypothesis increases in the right side of the matrix, which represents the least contributing heads, thereby implying that the sample data is more homogeneous for this portion of the matrix. We also note that the head-wise pattern is more discernible than the sector-wise pattern. Accordingly, Table 6 further s ummarises the occurrences of acceptance of null hypothesis in absolute and percentage terms for each of the FBT heads. The heads has been listed in decreasing order of their contribution in total collection. The table has been horizontally divided into two parts, thus listing the top 10 and bottom nine heads separately.

Now, two distinct patterns are clearly discernable. In the case of the top 10 heads, the occurrences of acceptance of null hypothesis are mostly between 0% and 15% except for the head T elephone. For this group of top 10 heads, overall, the null hypothesis has been accepted in 20 out of 220 (22 sectors × top 10 heads) instances of tests which give an acceptance of 9%. For the heads conveyance, gift and maintenance of car, the null hypothesis has not been accepted even once. This shows that for these heads of expenses, the data of individual taxpayers are very heterogeneous.

It is easy to notice that in case of the bottom nine heads, the occurrences of acceptance of null hypothesis have suddenly increased and are in the range of 30% to 70% for all the heads except for hospitality and rep, runn, dep on aircraft. O verall, for the bottom nine heads, 180 chi-square tests have been conducted out of which, in 79 instances (43%), null hypothesis of equality/homogeneity of sample data has been accepted.

Economic Political Weekly

EPW

Table 6: Chi-Square Test: Occurrence of Acceptance of Null Hypothesis for FBT Heads

*There are only 15 economy sectors for which this head has shown any collection and in none of the instances, the null hypothesis has been accepted. ** There are only 11 economy sectors for which this head has shown any collection and out of these 11 tests, in 3 instances, null hypothesis has been accepted. @ Contribution up to Rs 1 lakh per employee per year exempt.

If we analyse the distribution of acceptance of null hypothesis for different sectors of the economy, it is seen that there are some sectors where the sample data has been found to be more h omogeneous, for example in case of minerals-metals, steel, chemical-fertiliser, and media-entertainment sectors, where null hypothesis have been accepted in eight to 11 instances of test out of 19 tests, i e, about 50% times. For two sectors, namely, banking and infotech-ITES; the null hypothesis has not been accepted even once, showing high heterogeneity of sample data for these sectors.

However, it is the distribution of test results as per the heads of FBT which throws some interesting results. We have seen that for top 10 heads, sample data are not homogeneous whereas for the bottom nine heads, sample data are more h omogeneous. Table 6 also shows the valuation base for each head of expense. It is also easy to note that in case of heads with 50% valuation base, most of which form the bottom five heads, occurrences of acceptance of null hypothesis are significantly more. Similarly, the top 10 heads, most of which have 20% v aluation base, are most heterogeneous. The type of heterogeneity shown in the test by the top 10 heads is difficult to explain only on the basis of internal diversity and differences in individual organisations/taxpayers. What can we infer form this analysis?

It is natural for the taxpayers to attempt reducing their FBT liability, if that is possible within the four corners of law. Due to the available discretion for booking expenses under

d ifferent heads, it is logical to expect that taxpayers would be motivated to book more and more expenses under those heads where the valuation base for FBT is lower. From the results of the chi-square test, this indeed appears to be the case. Taxpayers are perhaps taking benefits of absence of any standardised method for c lassification and booking of expenses. This is the reason data for heads with a lower base are the most heterogeneous representing arbitrary booking practices and are the largest contributing heads. Such a p ractice is not illegal and to some extent natural also since it is always possible that some particular expenses have the p ossibility of being included in more than one category. On the same logic, generally the heads with a higher valuation base are the least contributing and the individual sample data is also more homogeneous because these are not experiencing any arbitrary booking. It should also be noted that an heterogeneity/arbitrary pattern emerges due to the fact that not all taxpayers would be indulging in cross booking of expenses and that the cross booking/shifting of expense would be quite random.

Further, perhaps due to this reason, the head employee welfare which is a wider category and which can accommodate a large number of other expenses like gifts, other benefits, scholarships, etc, is the largest contributing head for FBT. Other heads in the top 10, like “conveyance”, “sales promotion”, “tour and travel” are also general in nature making it possible to book different kind of expenses under them and therefore, show high variability in booking practices.

The heads which are least contributing like gift, scholarships, health club, etc, are specific in nature and it is difficult and too blatant to accommodate and book other expenses under these heads, though it is easy and perfectly legal to book such expenses under employee welfare. Further, there is no benefit of reduced tax liability by booking other expenses under these heads having a 50% valuation base. Similarly, the head t elephone is quite conspicuous and specific and it would be very blatant to book other expenses under this head which is the r eason for its high h omogeneity, although it has a low valuation base of 20%.

On the basis of above analysis, one is tempted to conclude that perhaps there are deliberate booking of expenses in such a way to reduce the FBT liability by business organisations, which is being reflected in higher heterogeneity in sample data in case of wider FBT heads having 20% valuation base.

7 Conclusions

What has been done through statistical tests can be termed as an introductory data mining and the findings are only preliminary in nature. Although, at the aggregate level, the collection pattern has shown stability and homogeneity, at the level of individual taxpayers, the data is showing a high level of heterogeneity. It only indicates that there may be attempts by individual taxpayers to book expenses in such a way which reduces their total FBT liability. There is a vast opportunity to further dig into data to gain a deeper insight. However, one suggestion which could be made on the basis of the above analysis is to r ecommend a uniform valuation base for all heads of expenses to remove the opportunity of tax avoidance. However, it would bring into question the very basic logic of bring FBT in the present from as the valuation bases are stated to be decided taking into consideration the nature of the expense head. F urther, recommending removal of the least contributing heads are not as easy because it may lead to shifting in the expense booking pattern to avoid FBT, thereby leading to a significant loss of revenue.

Some alternatives of FBT have been suggested like imposition of a flat rate of surcharge on corporate tax or making valuation rules of perquisite more comprehensive and incorporating FBT provisions in them, thereby only taxing employees and not the employers. All this requires further deliberation, informed discussion and empirical analysis of FBT collection data through involvement of all the stakeholders. Only then will it help in reform of the FBT regime and an overall i mprovement of tax policy formulation and taxation structure of our economy.

Notes

(n1 p1 + n2 p2 ) 1 We define p = —————— where p is com(n1 + n2)

bined population proportion, n1 = No of observation in 2006-07 = 350 and n2 = No of observations in 2005-06 = 350. The sample standard deviation S is given by

((

1 n 12)S = √ p 1–p) — + — .

n1

(p1 – p2) We calculate Z statistics as Z = ——— . S

The Z statistic so calculated is then compared to the critical value of Z statistics. The critical value for a given level of confidence (it has been taken at 5% in this case) is found by looking at the normal distribution table. It is a two-tailed test of hypothesis and the corresponding critical value which leaves 5% area of the Standard Normal Distribution in each of its tails (thus keeping 90% area within the acceptable limits of critical value) is 1.645. Thus, if we have value of test statistics below the critical value (1.645), we accept the null hypothesis and if it is more than critical value, the null hypothesis is rejected.

2 Mathematically, H: p1i = p2i for all i

H1: At least one i not the same, where p1i is the proportion of FBT collection for i th head for the year 2006-07, and p2i is the proportion of FBT collection for the i th head for the year 2005-06, there are in total 17 heads, i e, i varies from 1 to 17. Chi-Square (X2) test Statistics is calculated as

with (n-1) degrees of freedom, where n is total number of FBT categories, i e, 17, and the degree of freedom, thus, is 16. The X2 statistics so calculated is compared from the critical value of chi-square distribution for given degree of freedom and confidence level.

3 “Per cent contribution” here means per cent contribution by a FBT head as a per cent of total FBT collection for a particular taxpayer. It has also been called “proportion of FBT collection” in subsequent paragraphs.

4 A detailed discussion of ANOVA is too complicated to be elaborated here. It would be sufficient to note that the ANOVA conducted in this case has indicated a strong interaction effect, which directed towards conducting one-way ANOVA or separate chi-square tests.

5 Mathematically, H: p1 = p2 = p3 = …. = p

on

: At least one p not the same. Where pis the

H1nn

proportion of FBT collection from n th FBT payer in the sample for a particular combination of sector and head; and n is total number of FBT payers in the sample representing a particular combination of sector and head.

Reference

Kishore, Praveen (2008): “Analysis of Fringe Benefits Tax and Its Collection Pattern”, Economic Political Weekly, Mumbai, Vol XLIII, No 33, Issue 16-22 August, p 41.

Economic Political Weekly

EPW

january 3, 2009