With his recent provocative article "Earn poor, spend rich", Surjit Bhalla has initiated an important debate on whether the Indian tax-GDP ratio is lower than what it should be or whether it is "just right".

In fact, it was Rakesh Mohan who forcefully argued last January at the conference organised jointly by the NIPFP and the IMF that undertaxation is at the root of the Indian fiscal problems.

In the conference, Indira Rajaraman also argued the point in her paper, "Fiscal Developments and Outlook in India."

Moreover, at the recent NCAER-NBER conference at Neemrana, I also presented evidence of undertaxation based on the data collected from the IMF's Government Finance Statistics. This is an important issue and the public and the policymakers need to know whether and to what extent Indians are undertaxed.

Nevertheless, the tax ratios prevailing in other countries can serve as a useful indicator and therefore, international comparison does provide some guidance on the tax burden, at least on a comparative basis.

Bhalla's analysis shows that at 15 per cent, the tax-GDP ratio in India is just right for the level of per capita GDP it has, and the problem is with expenditures.

He establishes his case by regressing the 1999 tax-GDP ratios in 82 countries on their per capita PPP GDP in 1993 and its square term.

To get this sample, he puts together information from the GFS of the IMF, the World Bank sources, and the OECD. Unfortunately, he does not give his regression results and therefore, it is not possible to ascertain the significance levels of the coefficients.

There are a number of conceptual problems with Bhalla's estimates and conclusions about the level of undertaxation cannot be established unless these are resolved. The first refers to whether it is appropriate to take per capita PPP GDP or simply dollar-converted per capita GDP.

The second relates to the form of the function to be used for regression. Why should one choose to take the log of the dependent variable and not simply estimate a linear function?

Further, even after taking the log values why should one take the square term of log per capita PPP GDP as an additional explanatory variable?

Even if one hypothesises an exponential relationship between tax ratio and per capita income, why should one expect even that relationship to be non-linear? Do the statistical estimates show that putting the log term of per capita income and its square term improves the statistical fit?

An equally important conceptual question is regressing the1999 tax-GDP ratios on 1993 log per capita income PPP dollars. The fact that India's position in the predicted line simply implies that at the 1993 configuration of incomes, India's tax share is "just right".

The problem, however, is that although the tax ratio in 1999 remained at almost the same level as in 1993 (14.2 per cent) the per capita PPP GDP in India increased from $1,220 in 1993 to $2,149 in 1999. Surely, you cannot judge the tax-GDP ratio of India in 1999 at the 1993 per capita incomes.

Since the growth of per capita income in India during the last decade was much faster than the world growth rate, the changed configuration imparts a downward bias to Bhalla's estimates of predicted tax ratios.

I have independently estimated regressions to estimate the extent of undertaxation in India based on GFS data. Consistent data sets from the GFS are available for 2001 for 40 countries.

In fact, for comparability, it is not advisable to mix the data from different sources unless the concepts used are uniform.

For example, the data presented in the World Development Reports refer to only central government taxes. The advantage with the GFS statistics is that they are based on standardised definitions.

The results estimated for some of the regression functions are summarised in the Table. Of the various regression models estimated, the table reports only four equations.

These are the regressions of tax-GDP ratios of different countries on their per capita GDP dollar converted and per capita PPP GNP in linear and log terms.

Equations were estimated, also by taking the square terms of the independent variables, but in none of the cases were these coefficients found significant and, therefore, are not reported here.

The analysis shows that the linear equation with dollar-converted per capita GDP as the independent variable has the highest explanatory power and next, the dollar-converted per capita GDP in log terms.

The explanatory power of the equations when per capita PPP GDP is taken is actually lower. Interestingly, the fitted values of the tax-GDP ratio for the per capita GDP level in India in respect of every estimated equation are lower than the actual ratio in 2001-02 by 2.5 to 6 percentage points.

In fact, the difference is about 5-6 percentage points when estimates of linear equations are considered, whether dollar-converted GDP values or PPP GDP values are taken. In contrast, the difference was about 2.5 to 3 percentage points when log values of the independent variables are considered.

The available evidence, thus, shows that the tax-GDP ratio in India is lower than the level it should have for its per capita GDP by at least 2.5 per cent and the difference could be as high as 6 per cent if the best fitting equation is chosen.

It is, therefore, important to focus reform efforts to increase the tax ratio. Of course, this does not mean that the strategy to increase the tax ratio lies in increasing the tax rates. The strategy is to reiterate that tax administration is tax policy.

Nor does the emphasis on increasing the tax ratio dilute the need to focus on exercising economy in expenditures and improving delivery systems for public services.

Policies to both increase the tax ratio and contain expenditures are necessary to phase out fiscal imbalances, and improving delivery systems is necessary to improve standards of public services.

The author is director, National Institute of Public Finance and Policy