Recent developments in the Institute's domestic macromodel.

Wren-Lewis, Simon

RECENT DEVELOPMENTS IN THE INSTITUTE'S DOMESTIC MACROMODEL

The forecast published in this Review uses the latest vintage of the Institute's domestic econometric model, version 11.4. In comparison with the model described a year ago (Model 11.1, see Wren-Lewis, (1988)) some of the main developments are: 1) The key price and wage equations are both forward looking, and are based on explicit dynamic theories of adjustment. This is perhaps the first time that a large quarterly econometric model has attempted to identify the structural characteristics of nominal inertia in the UK. 2) Our econometric work has suggested that the real interest rate, working through the cost of stockholding, plays an important role in influencing prices. In the longer term, changes in real interest rates also influence the model's NAIRU by the same route. 3) The model contains a new disaggregation of output by industrial sector, and we have extended our vintage production technology to cover distribution and business services. 4) Unemployment is now modelled according to claimant status, rather than as part of a labour supply decision. Partly as a result, unemployment no longer appears to be the most useful measure of excess supply in the labour market when it comes to explaining earnings. 5) The new version of the model contains a detailed set of equations explaining the capital account, with econometric relationships for direct and portfolio investment inflows and outflows and net overseas liabilities of domestic banks. These are augmented by equations for other net capital flows--primarily official reserves. The full set of asset stocks permits a more sophisticated model of interest, profit and dividend receipts which are an important component of the overall current account.

In addition various changes have been made to the financial sector, as well as other areas of the model. We end this note with a short discussion of the effects of interest rates on the economy.

1. Forward looking prices and wages

One of the most quoted reasons why demand shocks have persistent effects on output is nominal inertia. It may take a considerable time, perhaps years, before wages and prices adjust to `crowd out' a nominal or real demand shock. This is certainly a feature of earlier versions of the Institute's model; Joyce and Wren-Lewis (1989A) show, for example, that in Model 11.1 it takes about five years for the model to settle down to its NAIRU after a fiscal policy change.

Despite the apparent importance of price rigidity in most econometric macromodels, there has been surprisingly little formal econometric modelling of the factors behind it. Typically macromodels have `let the data determine the dynamics' in price and wage equations, and have not attempted to impose any theoretical structure on it. In Joyce and Wren-Lewis (1989B) a model for manufacturing wholesale prices is estimated in which a combination of reputation costs (based on the idea that large price changes are most easily perceived) and asynchronised pricing leads to nominal inertia. This implies not only that prices adjust gradually, but also that firms attempt to anticipate changes in costs and demand. Prices depend not only on lagged prices, but also on expected wages, productivity, import prices, etc.

In Model 11.4 this structure is also applied to consumer prices. In each case the degree of nominal inertia is summarised by two parameters; the relative importance of reputation costs and the degree of asynchronisation. These changes to the price equations represent in many ways a natural extension of the forward-looking employment, stockbuilding and investment equations first introduced in Model 8 (see Hall and Henry (1985)).

A more familiar element of nominal inertia is the effect of wage contracts (see Taylor, 1980). Once again, inertia coupled with rational optimising behaviour implies agents look forward. In Moghadam and Wren-Lewis (1989) results are presented which suggest that labour market participants may well form rational expectations about future prices when fixing nominal wages. This forward-looking behaviour was already a feature in earlier versions of the model. In 11.4 it is extended so that agents attempt to anticipate other factors influencing the wage bargain, such as productivity changes and unemployment.

The introduction of a complete set of explicit forward expectations terms into the key price and wage equations has had an important effect on model properties. (It has also increased the number of expectations variables in the model, which now number over 90.) In general, wage-price dynamics are faster. The most noticeable differences arise in two areas. Firstly anticipated changes now influence prices and wages before they occur, even if the exchange rate is unchanged. Secondly, because earnings depend on expected unemployment, changes in activity (anticipated or unanticipated) have a quicker effect on real wages. This greatly speeds up the transmission of real shocks into the inflationary process. In earlier models, when wages depended on lagged unemployment, real shocks had a delayed effect on wages because employment responded gradually to output movements. Now, however, because unemployment movements are anticipated, changes in output influence real wages much more quickly.

Chart 1 shows the effects of a rise in government expenditure (worth about 1/2 per cent of GDP) on GDP and the price level (consumer prices) in the new model. The initial increase in GDP is amplified partly because the exchange rate drops by about 3 percentage points, anticipating future current account deficits. However, prices respond fairly rapidly to both the lower exchange rate and higher activity, so that after two years any gain in competitiveness is eroded. This helps crowd out much of the initial expansion in GDP. In the long run, GDP remains higher because the switch from private to public expenditure reduces imports, which in turn allows the real exchange rate to appreciate in the long run to restore current account equilibrium. An appreciation in the real exchange rate helps counteract the effects of lower unemployment on wage pressure, so unemployment is still around 40,000 lower after about six years. In terms of GDP, the model comes close to its new equilibrium rather sooner (about 3/4 years) than the 5/6 year adjustment period suggested in Model 11.1 (see Joyce and Wren-Lewis, 1989A). This shows that the move to forward-looking wage and price equations has speeded up the price adjustment process, but also that nominal inertia remains very important for short-run analysis.

2. Real interest rates and the NAIRU

Joyce and Wren-Lewis (1989B) also examine the long-run determination of manufacturing prices. An interesting new result is that the cost of stockholding is found to have a large effect on price setting. The theoretical reason why this might be the case is that manufacturing firms hold large stocks of both finished goods and materials (currently worth about 50 per cent of annual output), and these stocks respond permanently to changes in output. As a result they are an important element of marginal costs. The cost of stockholding variable is currently equivalent to the expected real interest rate, although there have been important changes to the tax regime in this respect over the last two decades (see Darby, 1988). This variable also appears to be an important determinant of consumer prices and wages.

The long-run structure of the wage and price equations interact to help determine the model's NAIRU; the level of demand pressure in the goods and labour markets that is consistent with steady inflation. This interaction is described in Joyce and Wren-Lewis (1989A), and its relationship to other aspects of the model's supply side is discussed in Wren-Lewis (1989). As increases in interest rates push up prices, they also tend to reduce the long-run level of output, although their effect is partially offset in the earnings equation.

Our long-run model of earnings behaviour has also changed slightly in other ways. We continue to assume that productivity increases are fully passed on into higher earnings, but the equation now gives a higher weight to manufacturing productivity compared to productivity in the rest of the economy. This helps explain recent strong growth in real earnings. (For a discussion of factors behind recent manufacturing productivity growth, see National Institute Economic Review, no. 128). We did not have any success in including structural measures of regional mismatch, but an industrial mismatch variable continues to have a strong influence. A rise in real import prices now reduces real earnings slightly, partially offsetting the effect of real exchange-rate changes on the Model's NAIRU. Finally we now find that the effect of a rise in employers' national insurance contributions on labour costs is partially offset by lower earnings, but this is not the case for other non-wage labour costs. (It must be said, however, that these and other tax effects on earnings were not very robust to specification changes.) One final change, to the unemployment term, is discussed below.

3. Output, employment and investment by industrial sector The model now splits GDP into eight output categories: manufacturing (SIC2-4), construction (SIC5), oil (SIC13), distribution (SIC6), business services (SIC8 less SIC85), public services (SIC9 and 00), imputed rent (SIC85) and a residual category, which is largely transport, communication and non-oil energy supply (see Box 1). These output categories (excluding oil and rent) are explained through a matrix of estimated coefficients related to expenditure components, in such a way that none of the output categories is a `residual'. Differential movements in expenditure categories explain a large part of variations in disaggregated output, but it does not explain the full extent of the steady rise in business services (and the associated decline in manufacturing) over the last decade.(1)

Our vintage model of the manufacturing production process, and its associated models of capacity, employment and investment, are described in Darby, Minford and Wren-Lewis (1989) and were incorporated into Model 11.1. In the latest model this system has been extended to two of the new output categories; distribution and business services. As a result, new investment in either sector can directly increase labour productivity in that sector. For example, a 10 per cent rise in investment over a three year period will raise labour productivity after five years by 1 1/2 per cent, 1 per cent and 1/2 per cent in the manufacturing, distribution and business services sectors respectively. New equations for investment in each sector suggest important influences from real share prices and the post tax user `cost of capital' alongside movements in output.

4. Unemployment Traditionally economists have seen unemployment primarily as a measure of labour supply. However recent changes in definition have undermined the validity of this association. Gregg (1989) argues that unemployment is best modelled by analysing the claimant status of particular groups, and that this claimant status may in many cases have little to do with a desire to find work. As a result, the relationship between employment and unemployment changes depends inter alia on whether the change in employment occurs among males or females, and whether workers are full or part time. Model 11.4 incorporates new total and short-term unemployment equations based on this work. Changes in the industrial structure of employment influence the level of unemployment in the model through their effects on female and part time employment.

Viewing measured unemployment as reflecting claimant status rather than labour supply factors also calls into question the suitability of unemployment as a measure of excess supply in the labour market influencing wages. Although unemployment may contain some information about labour supply decisions, which is relevant to wage pressure, it may also change because of movements in groups with different claimant status which may have no implication for wage pressure. (This argument also applies to short-term unemployment.) In our latest work on the long-term determinants of real earnings, we experimented with an alternative, broader measure of excess supply that represented the percentage of the population between 18 and 59F/64M not in employment. This alternative measure explained more of the movement of real earnings than claimant unemployment, and so it replaces short-term unemployment in Model 11.4's wage equation. It has also increased substantially the impact of employment changes on real earnings. In addition, changes in the composition of employment by sex or industry have a long-run effect on claimant unemployment but no long-run influence on real wages. (There is an important short-term impact through an industrial mismatch term, however.)

5. International capital flows The Model now contains equations determining both inward and outward capital flows disaggregated in terms of direct and portfolio investment. The direct investment equations are based on the work of Pain (1989), and take a stock adjustment form. They imply the real factor prices (labour and capital) in the UK and overseas have a strong influence on direct investment flows. Outward flows also respond to the level and change in world activity, and there are dynamic effects from the growth of exports and real corporate profits. Inflows move in line with domestic output, and there are additional effects from oil prices and world output. Increases in domestic real unit labour costs are found to be a significant investment disincentive.

The equations for portfolio flows are obtained from stock equations of a conventional type, deriving the share of the portfolio in total wealth as a function of relative returns at home and abroad. Returns include coupon rates on bonds, dividend yields, capital gains and expected exchange-rate revaluations.

These equations together with equations for net flows within the banking sector form the basis of a structural model of the exchange rate in Westaway and Pain (1989). However this work was explicitly preliminary, and so in Model 11.4 the exchange-rate equation introduced with Model 11.1 is retained. The capital flow equations have been utilised, however, to improve the modelling of overseas IPD flows (see below), and the net acquisition of non-liquid financial assets by the corporate sector.

6. Financial variables Long-term interest rates should be related to expected short-term rates. If future short-term interest rates are expected to decline, other things being equal current long rates will be below current short rates. Our new equation for long-term interest rates is now based on this simple term structure theory. As a result, temporary increases in short rates will have a smaller impact on long rates than more permanent changes.

Long-term interest rates have an important influence on a number of areas of the model, including share prices. A new equation for nominal share prices also contains a short-run influence for world share prices in sterling terms, and imposes a long-run unit elasticity with respect to the sum of terms in dividends and profitability. Although this equation is not itself forward looking, it does depend on two other `jump variables'; the long rate and the exchange rate. (See Westaway and Pain, 1989, for further details.)

We have also replaced our equations for M0 and M1 with a new model of the demand for cash, based on the work reported in Walton and Westaway (1988). This attempts to endogenise the influence of financial innovation by including a term in the cumulated level of post-tax interest rates. The intuition behind this variable is that changes in interest rates give investors an incentive to adopt new cash management techniques, thereby giving rise to `ratchet' effects in the demand for cash.

7. Other developments We can briefly list some of the other changes embodied in the latest domestic model. Full details can be found in a complete Model Manual available from the Institute. A PC version of the model is also available for forecasting and simulation analysis. (i) A new equation for manufacturing exports contains a `stochastic trend', which suggests that the historic tendency for the UK to lose export market share at a constant level of competitiveness has come to an end in the last few years. This trend replaces a term in export profitability (see Anderton (1989) for further details). (ii) Most of the IPD equations on the model have been revised, to reflect new information provided by the capital flow equations. Public sector net property income is now the residual category in this system. (iii) The oil sector of the model has been substantially simplified. (iv) New equations for stockbuilding utilise the new output disaggregation, and introduce a cost of stockbuilding term in other stocks.

8. The response to interest-rate increases The latest version of the model suggests that the impact of higher interest rates depends crucially on how long interest rates stay higher. Chart 2 plots the impact on GDP of a 2 point increase in interest rates lasting either for one year or for two years. The two-year rise has a much bigger impact on GDP in the first year for two reasons. The first is that the size of the initial exchange-rate appreciation will be larger if interest rates are expected to stay higher for longer. The second is that long rates will also rise by more initially if the increase in short rates is expected to be more prolonged (see 6 above). Both long rates and the exchange rate have an important impact on share prices, which fall initially by about 1 per cent and 2 per cent in the two simulations. Share prices in turn have important effects on both investment and wealth in the model.

The difference in the impact on inflation in the two simulations is even more noticeable. If interest rates are expected to rise for only one year, the reduction in price inflation after a year is very small. This is because the effects of lower activity and the exchange-rate appreciation are partially offset by the direct costs of higher interest rates incurred by firms (see the discussion of the costs of stockholding effects in 2. above). These effects have less influence on wages, however, which fall by about 1/2 per cent after a year. When the interest-rate rise lasts for two years, price inflation falls by more, about 1/2 per cent, after a year. The increase in costs caused by the rise in interest rates remains the same, but the deflationary effects through activity and the exchange rate are now greater. Needless to say the effects of a three or four year interest-rate increase on activity or inflation would be greater still. This suggests that to be successful the current high interest-rate policy needs to be perceived as more than very short term. (1)There has been an increasing tendency for manufacturers to contract out various service functions, leading to a switch between industrial sectors, but we have not as yet attempted to model this.