Considerable attention has been devoted to the reform of the tax and benefit structure, both in the UK and elsewhere. The impact of such reforms seems relatively easy to predict, at least if we ignore its potential effects on the behaviour of the agents in the economy. However, growing concern with the behavioural implications of the tax system, in particular its effect on incentives to work, has led a number of researchers to analyze these reforms more closely and in a more comprehensive framework.

In order to highlight the present state of the art in this important area of research and examine its possible implications for tax policy, CEPR organized a workshop on Micro Data, Micro Models and Tax Policy Analysis on 2 May. Mervyn King (LSE and CEPR) discussed research carried out with his colleagues at LSE, and Richard Blundell (UCL and CEPR) outlined work he had undertaken with colleagues at UCL, the Institute for Fiscal Studies and Manchester; these represent two of the leading UK research teams in the subject. The meeting was chaired by John Muellbauer (Oxford and CEPR).

Mervyn King began by emphasizing that the enormous diversity of households meant that the effects of a particular tax reform could vary greatly across households. If policy conclusions were to be drawn from this analysis, King argued, they must not be overly sensitive to the particular assumptions made by the researcher. In addition, analysis of tax and benefit reforms should satisfy three other requirements. Estimation and selection of models should take into account the objectives of the policy-maker and not merely those criteria suggested by statistical or econometric theory. Analysis should explore fully the distributional consequences of a reform, analyzing which households gain and which households lose, instead of just looking at 'deadweight loss', which simply provides a measure of gains and losses averaged over all households. It should be possible to simulate the effects of tax and benefit reforms using a package which is easy-to-use and widely available.

The first step in analysing the behavioural and welfare consequences of tax and benefit reforms is to construct a model of households' labour supply, i.e. the relationship between the hours of work chosen, the marginal after-tax wage, and the level of other after-tax income. The comprehensive character and relatively high quality of the British Family Expenditure Survey (FES) clearly suggests using the household survey data it contains in order to estimate the labour supply function.

Modelling labour supply does involve a trade-off, between the 'flexibility' and generality of the functional form chosen and the ease with which we can calculate a valid measure of welfare for this functional form. A more flexible functional form can better capture the salient features of the data, but its generality may make it difficult to calculate the 'virtual wage', i.e. the wage at which an optimizing agent would choose a specified decision. These virtual wages are required for the calculation of welfare levels for each household.

To illustrate the main features of his approach, King analysed the effect of a change in employees' national insurance (NI) contribution on labour supply and welfare, using his program 'Microtrap', available for microcomputers. He assumed a simple linear labour supply function estimated using FES data for some 3000 households. He simulated the 1985 increase of 8.4% in national insurance contributions. The change led (in the simulation) to an average decrease in the number of hours worked, a welfare gain on average and a 4% decrease in government revenues.

A full distributional analysis suggested that the greatest welfare gains occurred in the households with highest pre-reform welfare. Labour supply was reduced most sharply among 'midrange' households. King noted that policy conclusions drawn from the simulation were conditional on the specific sample used, and on the various assumptions made, in particular the functional form of the labour supply equation.

In the second presentation to the workshop, Richard Blundell emphasized the important, but sometimes overlooked, fact that models which are too simple can seriously prejudice the outcome of our analyses. Blundell based the discussion on his UCL Discussion Paper, 'Alternative Specifications of Labour Supply and the Simulation of Tax Reforms' (with Costas Meghir, Elizabeth Symons and Ian Walker). In this paper a variety of different labour supply specifications had been proposed, ranging from general and complex to simple and restrictive models, reflecting different degress of flexibility in the underlying substitution possibilities between leisure and goods. In particular, economic theory might suggest that for workers at low wage levels, an increased wage rate could induce them to work more hours; workers earning higher income might choose to take more leisure if wage rates were increased. Such behaviour can only be accommodated within a more flexible model specification and can be crucial for policy conclusions. Of course, the trade-off between flexibility and computational feasibility mentioned earlier must also be taken into account.

As part of their research Blundell and his associates had developed, in collaboration with the Institute for Fiscal Studies, an interactive package designed to simulate the effects of tax and benefit reforms. Ian Walker (Manchester and CEPR) demonstrated the package using an IBM Personal Computer.

For the purposes of the demonstration, the simulation was based on a limited sample of 50 households. The UCL Discussion Paper, however, contains a more comprehensive analysis of two tax reforms using a sample of 2195 households, from the 1981 FES, each containing two married adults of working age with the husband an employee. As male labour supply seems to be insensitive to tax changes, attention was focussed on the married women, whose behaviour is thought to be more sensitive to such changes. Reform A involved reducing the married man's tax allowance (MMA), i.e. the highest taxable income free of income tax, to the level of its equivalent for married women, the single person's allowance (SA), which is at present 36% less than the MMA. In addition, child benefit was increased by 115%. Reform B, on the other hand, involved a 50% rise in both the MMA and the SA, the abolition of the national insurance (NI) system and a substantial increase in the standard tax rate from 30% to 45%.

Whereas reform A implies mainly income effects, reform B contains both income and wage effects, making it harder to predict its effects on labour supply. It is interesting to note that for both reforms, model specification becomes crucial when both income and wage effects are involved.

For reform A the behavioural responses do not differ dramatically among the model specifications considered, but government revenue does vary. In the case of reform B, government revenue is not sensitive to the assumed model specification, but labour supply responses are quite different. Reform A leads to an increase in labour supply for those who work large numbers of hours, as they typically have no children. Women working fewer hours tend to have children, and two of the four models suggest a decrease of hours worked. As both wage and income effects are present here, however, the models differ in this region, with its increased marginal tax rate. Reform B causes those groups working relatively many hours to reduce their labour supply.

Blundell argued that these simulations demonstrated the benefits of using flexible model specifications. These benefits clearly outweighed the computational convenience of simulating with simpler, but potentially misleading, model specifications.

To conclude the session, a stimulating discussion covered various topics, including analysing rationed purchases, deliberate misreporting in the survey data analysed, the treatment of those households that do not respond to the survey, as well as lags present in the data.

The presentations to the workshop demonstrated that careful consideration of labour market incentives is essential in the analysis of reforms to the tax and benefit structure, especially if we are concerned with the distributional effects of tax policy. As individuals will only change their labour supply if it is beneficial for them to do so, failing to take into account these incentive effects will understate the gains and overstate the losses attributable to the tax reform in question. Therefore any analysis ignoring behavioural effects will implicitly be biased in favour of the status quo, as such an analysis will not correctly evaluate the merits of the proposed reform.

The existence of the user-friendly and widely available simulation software demonstrated at the workshop suggests that the work of these research teams will assume increasing importance in the formulation of tax and benefit polices.