|
|
Anglo/Finnish
Workshop
Macroeconomics
and Capital Markets
One of the goals of
the Centre is to bring together its Research Fellows with their
colleagues from other countries in order to exchange ideas and to foster
collaborative research initiatives. Financial support from the Yrjo
Jahnsson Foundation enable CEPR Research Fellows to meet academic and
government economists from Finland at a two-day workshop last December
at CEPR.
The workshop was opened by Seppo Honkapohja (Yrjo Jahnsson
Foundation) with a paper entitled 'Speculation, Instability, and
Government Policy', which described the progress of his research on the
dynamic aspects of government policy, especially the relationship
between the actions of the government and those of the private sector.
Individuals will adjust the allocation of their wealth between public
and private sector assets in response to current government policy. But
since individuals must also plan for the future, their current portfolio
choice will also depend on their expectations of future government
policies. Since these policies are uncertain, agents in the private
sector will assign different probabilities to each policy option:
essentially, they speculate against different future policies.
Honkapohja's research was designed to explore how such speculation will
in turn affect current government policy.
In Honkapohja's main model, the government is assumed to hire capital
goods from households, which it uses to provide free public services.
Households allocate their stocks of capital goods between the public and
private sectors, aiming to maximise their expected utility over their
lifetime. They must therefore allow for a changing pattern of taxes and
subsidies on their incomes over time. The rentals from the public sector
are used by the private sector as money balances, holdings of which vary
with total expenditure on goods produced by the private sector (since
the public sector by assumption provides goods and services free of
charge).
The allocation problems facing households and the policy problems facing
the government in this model raise difficult technical issues, which
Honkapohja could only solve using some simplifying assumptions. One of
these was that uncertainty concerning future tax changes could be
described by a Poisson probability distribution. This would be
appropriate if the government only changed tax rates when its objectives
moved outside a band of target values. Within this 'band', the costs of
a change in policy would outweigh its benefits.
The participants recognised the technical problems raised in
Honkapohja's analysis. Marcus Miller (Warwick and CEPR) suggested
an alternative approach, in which the probability of falling outside the
target band at any given time would depend on the position of the
economy. Mike Wickens (Southampton and CEPR) wondered why a
government should respond to a one-off shock, as it did in Honkapohja's
analysis. Policy changes did not appear in practice to be random, and
subsequent discussion considered allowing the probability of a change in
policy to depend on the level of money holdings or the inflation rate. David
Begg (Worcester College, Oxford, and CEPR) drew attention to the
technical problems this approach would raise, and participants
considered how these problems might be resolved.
Economic theory often implies constraints on the parameters of a set of
behavioural equations, such as the homogeneity and symmetry constraints
in demand systems. To obtain parameter estimates consistent with the
underlying economic theory, these constraints should be imposed in
estimation. Alternatively, one may want to test the constraints by
estimating the model in both constrained and unconstrained form. There
are, however, reasons why these constraints might not hold exactly. The
data used in demand or cost studies is often aggregated and the
constraints on the behaviour of an individual agent may be invalid at an
aggregate level. Hence, testing the constraints or imposing them exactly
may be inappropriate for aggregate data.
In addition, the optimizing behaviour on which the derivation of the
parameter constraints is based may be imperfect. Such optimisation
errors are a common justification for adding an error term in
behaviorual equations. These arguments suggest that one might wish to
treat the parameter constraints implied by economic theory as
stochastic. Such an approach would also allow an assessment of the
sensitivity of the results to imposition of the constraints.
Pekka Ilmakunnas (Research Institute of the Finnish Economy)
explored this approach in his paper 'Stochastic Constraints on Cost
Function Parameters: Mixed and Hierarchical Approaches'. He considered
two ways of incorporating randomness in the estimation of cost functions
(which are defined as the minimum cost to the firm of producing at a
given output level). The first approach consisted of simply adding a
random term to the constraints themselves; so-called 'mixed' estimation
procedures were appropriate in this case. The second approach assumed
that the constraints were satisfied exactly, but treated the parameters
themselves as stochastic, giving rise to a 'hierarchical' approach. The
choice between the two approaches will depend on how the randomness is
assumed to arise: if the parameters are stochastic, the hierarchical
approach is better; if not, the mixed approach is preferable.
Ilmakunnas applied these approaches to a model of structural change in
Japanese industry and to Berndt and Wood's study of the demand for
energy in American industry. He found that the results are changed
markedly by introducing uncertainty into the constraints, showing the
potential importance of the approach and giving a further illustration
of the fragility of econometric results. He noted, however, that if the
constraints or the parameters are not subject to errors, the use of
either of the two methods will result in worse estimates.
This last point prompted Richard Blundell (University College,
London, and CEPR) to emphasise the importance of testing for random
parameters before estimation is undertaken: several such tests are
available in econometric literature. Blundell also argued that these
tests should be carried out against plausible and interesting
alternative hypotheses and should not be treated as casual
afterthoughts. Christopher Bliss (Nuffield College, Oxford, and
CEPR) pointed out that the aggregate quarterly data used by Ilmakunnas
might be inadequate to test hypotheses derived at the level of the
individual agent. Begg and Wickens also drew attention to the need for
an appropriate theoretical framework: factors such as habit formation
and costly adjustment might also be important. In the study of technical
change that Ilmakunnas discussed, a gradual learning process was
probably in operation, and his methods may have attributed the effects
of learning to errors of optimisation or aggregation.
In 'Monetary Stabilisation Policy in an Open Economy', Marcus Miller
(Warwick and CEPR) analysed the design of monetary stabilisation policy
when a floating exchange rate and sluggish domestic price adjustment
cause the exchange rate and the real interest rate to respond to
disturbances more quickly than does the rest of the economy. The
objective of the government in Miller's model is to minimise the
weighted sum of the deviations of output and of the inflation rate from
their target values.
In a closed economy, the optimal anti-inflation policy involves reducing
output; once inflation has been eliminated, output can be increased.
Difficulties arise with such a policy in an open economy, however: a
flexible exchange rate, through its effect on import prices, will
influence the inflation rate. Arbitrage in capital markets means that
the exchange rate will in turn depend on the expected differential
between domestic interest rates and those in the rest of the world. Core
inflation will therefore reflect sophisticated forward-looking
behaviour, and policy- making is complicated by the effect of monetary
policy on the exchange rate.
Miller examined three different policy rules. In the first, the policy
makers choose the path for interest rates which is optimal at each
moment in time. This policy may well be 'time- inconsistent': the
interest rate that the policy maker announces today as the plan for some
future time may not be optimal when that time arrives. Future interest
rates may be different from those announced in plans today even though
the economic circumstances which gave rise to the original plan remain
unchanged. The second policy considered by Miller is one which is
optimal, subject to the constraint that the government may not renege at
a later date on its current announcements of future policies; this
policy is termed 'time-consistent'. The third policy analysed by Miller
is one in which interest rates are set with no consideration of their
effect on the exchange rate.
In a simulation study, Miller found that the third, 'exchange- rate
exogenous' policy performed very well, outperforming the second policy
and nearly matching the first policy (in which policy makers do the best
they can in every period). Indeed in the simulations he conducted,
Miller detected a close resemblance between an optimal policy based on a
simple feedback rule, in which policy instruments respond to the state
of the economy, and the 'exchange-rate exogenous' policy, even though
this policy rule is not based on feedback.
David Currie (Queen Mary College, London, and CEPR) observed that
Miller's results implied that policy makers could make more gains from
allowing for the openness of the economy than they could from being able
to renege on their past commitments. Essentially this arises when
exchange rate appreciation allows the 'export' of domestic inflation,
but there is always the problem that other governments can adopt similar
tactics. Perhaps coordination would be better in such circumstances.
Currie also took issue with Miller's argument that many problems arise
because policy- makers attach too little weight to the future
implications of current policy. Although many problems would be resolved
if governments took full account of the future, Currie found it
unreasonable to assume they would do so. Matti Pohjola
(University of Helsinki) was interested in the 'game' played between
policy makers and private agents and wondered precisely how it should be
modelled, given that policy makers seemed to derive an advantage from
their ability to set policies and to influence private sector behaviour.
There was some discussion of this question and of the technical problems
involved in calculating the various optimal policies that Miller used.
Irving Fisher defined the real interest rate to be the nominal rate
minus the expected rate of inflation. Other things being equal, he
argued, the real rate should remain constant: it is the nominal rate
which should adjust as expectations vary. In his paper, 'Inflation,
Hedging and the Fisher Hypothesis', Matti Viren (Bank of Finland)
examined the behaviour of real interest rates in various countries over
the past 20 years. Viren used a proxy for expected inflation rates in
his regression analysis, which was designed to assess how far the real
rate has been constant. His approach allowed for a variety of random
shocks to the economy and for uncertainty regarding inflation. Viren
also sought to contrast the Fisher Hypothesis with a model of nominal
interest rates based upon 'hedging' by investors against inflation. In
this model interest rates contained a 'risk premium' to compensate
investors against the risk of capital loss owing to inflation.
Viren constructed his proxy for expected inflation as an average of past
actual rates, but this produced poor results in the regressions. The
coefficient on expected inflation should be approximately one if
Fisher's Hypothesis were correct, but Viren's estimates were much lower
than this. Furthermore, the results did not appear reliable for any of
the countries considered. The hedging hypothesis, on the other hand,
seemed to produce much better results and may provide a better
explanation of the behaviour of interest rates. But Viren's estimates
for this model also seemed rather unstable and unreliable.
Discussion of Viren's paper centred around the interpretation of the
Fisher Hypothesis, particularly the assumption 'all other things being
equal'. Marcus Miller called attention to recent theories which
attributed business cycle fluctuations to changes in patterns of
intertemporal substitution, which were caused by variations in the real
interest rate. Mervyn King (LSE and CEPR) suggested that any
allowance for inflation hedging and capital risk should play close
attention to the details of the tax system. Miller emphasised the
importance of the distinction between unanticipated and anticipated
inflation, particularly in short-run analysis.
The econometric problems involved in measuring expected inflation also
aroused interest. Richard Blundell thought Viren's use of a
backward-looking measure was a weakness. Mike Wickens suggested that the
actual rate was probably the best measure of the expected inflation rate
and that it should be used in the estimation process. David Begg was
concerned that Viren's estimates used the level of output and of the
money supply as explanatory variables, without allowing for the fact
that their values were determined jointly with the inflation rate. He
thought it important to explain inflation using variables that were not
themselves functions of inflation.
In 'Efficient Equilibrium in a Differential Game of Capitalism', Matti
Pohjola (University of Helsinki) sought to apply the methods of game
theory to investigate the potential efficiency of the basic capitalist
system. Workers and capitalists are the 'players' in Pohjola's game, and
each group tries to maximise their own welfare. Pohjola assumes that
workers determine wages and hence the distribution of income.
Capitalists control investment and hence the size of income. Under these
circumstances both groups can increase their own welfare by cooperating
with the other class. Without cooperation an 'inefficient' equilibrium
can result, since each class has an incentive to exploit the other
class.
Pohjola extends the model so that the game is played repeatedly, and
players can remember the past. Then an efficient outcome can be reached:
each player abides by the cooperative agreement, so long as neither
player has cheated in the past. Each player realizes the costs of
non-cooperation, so there is no incentive to deviate from the
cooperative rule. This shows that changing the information available to
the players will greatly change the outcome of the game, Pohjola
observed.
These results suggest that 'inefficiencies' such as unemployment and
slow growth are not intrinsic to capitalism. Pohjola conceded that the
structure of capitalism in his analysis was somewhat idealised. The
'classes' in Pohjola's model act as homogeneous groups, and if
inefficiency exists, it could be ascribed to one player's lack of
knowledge of the 'game'. Both 'players' or classes have perfect
information and are equally concerned about welfare in the future and in
the present.
Discussion of Pohjola's paper focussed on the technical problems
involved in modelling a game that lasts over many periods. Mark
Salmon (Warwick and CEPR) suggested that it should be possible to
find a variable that acted as a 'sufficient statistic' for the game,
incorporating all the relevant information concerning the past history
of the game. Using such a variable, Pohjola's repeated game could be
analysed as if it were a much simpler one- period game. Participants
agreed that game theory, when used properly, could usefully illuminate
many aspects of macroeconomics.
There is considerable debate as to whether the goal of full employment
is best pursued by means of tax cuts or increased government
expenditure. Neil Rankin (Queen Mary College, London, and CEPR)
used a 'disequilibrium' framework to analyse this question in his paper
'Taxation vs Spending as the Fiscal Instrument for Demand Management: A
Disequilibrium Welfare Approach' (available as CEPR Discussion Paper No.
84). Rankin based his analysis on a model that allows supply and demand
to differ in the goods and labour markets, although the money market is
assumed to be in equilibrium.
The results of such an analysis are sensitive to the nature of
government expenditure, and in order to assess this sensitivity, Rankin
considered three cases, in which government spending is treated as
'waste', a consumption good, or an investment good. For each of these
cases he calculated the 'optimal' policy, namely that which produces
full employment and the greatest increase in the utility of consumers in
the economy.
In Rankin's model increased government spending will increase welfare.
The optimal policy, however, will be to use tax cuts to achieve
full employment and government expenditure to achieve distributive
goals. When he assumed that the government is obliged to balance its
budget, however, matching expenditure to income, Rankin found that there
are some grounds for using government spending to achieve full
employment.
It is widely believed that the UK is at present experiencing 'Keynesian'
unemployment, in which supply exceeds demand in both the goods and
labour markets. Rankin found that if unemployment was Keynesian,
increased government expenditure will increase welfare in all three
cases.
Members of the workshop agreed with Rankin's approach but stressed the
need to analyse the institutional and political influences on government
spending and to incorporate monetary policy within the model.
The 'Ricardian Equivalence Theorem' has been the source of considerable
controversy in recent years. According to this theorem, the choice
between bond- and tax-financed government spending does not matter,
since an informed private sector realises that taxes will have to be
levied at some time to repay the increased expenditure and service the
bond issue, and it saves now to anticipate higher taxes in the future.
This means that if a government chooses to finance spending by issuing
bonds rather than by levying taxes, the bonds will not be treated as net
wealth by the private sector.
In 'Tax Cuts, Risk Sharing and Capital Market Implications', Erkki
Koskela (University of Helsinki) examined the impact of taxes on
private sector spending decisions. Koskela argued that the Ricardian
argument depended on a specialised and unrealistic set of assumptions,
and that under more plausible assumptions, the Ricardian equivalence
will fail to hold and the timing of taxes will have a Keynesian effect.
Koskela emphasised four difficulties with the Ricardian argument. The
first is uncertainty surrounding future income, which tends to reduce
current consumption. A reduction in current taxes will increase current
income, but disposable income must fall in the future to repay future
taxes. This expectation of increased taxation serves to decrease
uncertainty surrounding future income, so a shift from taxes to bonds
will cause current consumption to rise. Secondly, Koskela considered the
effect of distortionary taxes. A shift in these taxes from the present
to the future will have an ambiguous effect on current consumption.
Koskela also found that the Ricardian equivalence also collapses if
there is uncertainty about the future course of wages. Finally, he
allowed for the effect of credit market imperfections such as credit
rationing and differing rates for borrowing and lending. Such
imperfections also imply that the choice of taxes or bonds will have
real effects on the economy.
Matti Viren commented that the paper dovetailed neatly with other
recent work, most of which also tends to reject Ricardian equivalence.
Since a proportion of future taxes may be levied beyond the lifetime of
individuals in the private sector, it is important to allow for
bequests, and this casts further doubt on the theory. Colin Mayer
(St Anne's College, Oxford, and CEPR) suggested that the dates when
taxes are levied might affect total tax revenue, because of the
resultant differences in patterns of consumption. If so, it could be
difficult to foresee future tax levels. Ailsa Roell (LSE)
commented on other recent work, especially by King, that gave plausible
theoretical reasons for a divergence between borrowing and lending
rates.
In 'The Implication of State and Local Taxes on Corporate Policy', Vesa
Kanniainen (University of Helsinki) discussed a puzzling feature of
the Finnish tax system. It would appear that firms do not claim all the
corporate tax allowances they are permitted. Why is this? Kanniainen
noted that Finnish companies in fact face both national taxes such as
the profits tax, and local taxes which are based on the size of the
firm. There are also various tax allowances designed to increase
investment, and the firm must choose which ones it wishes to claim.
Reported profits may not match actual profits, although dividends must
be paid from the reported figure. Kanniainen argued that far from being
irrational, the failure to claim tax allowances represents optimal
behaviour, designed to reduce the impact of local taxes. Kanniainen used
a formal model to demonstrate this, and he then used a dynamic model to
examine whether this behaviour was also optimal over time.
The workshop generally agreed that Kanniainen's explanation of the
paradox was probably correct, but Colin Mayer wondered whether the
analysis could accommodate the effects of corporate finance and credit
markets.
Colin Mayer (St Anne's College, Oxford, and CEPR) presented the
final paper of the workshop, entitled 'Company Expectations and New
Information: An Application of Kalman Filtering' (available as
Discussion Paper No. 62). This joint work with Matthias Mors attempted
to model how UK companies formed their expectations concerning variables
such as sales and employment. Mayer and Mors took qualitative data from
CBI surveys giving the proportion of respondents who indicate that they
expect these variables to 'increase', 'decrease' or 'remain the same',
and he converted these data into a form which could be compared to the
realised values of the variables.
The 'Kalman filter' was designed by engineers as a statistical device to
combine uncertain observations on a variable with theoretical
information, to provide the best estimate of the actual state of the
variable. Mayer used this technique to develop simple models relating
firms' expectations to the past information available to them.
Expectations of orders and sales are difficult to model, yet even the
simplest statistical models constructed by Mayer tracked the actual
values of orders and sales considerably better than the data on
companies' own expectations. This suggested that firms do not use even
the simplest forecasting rules, although another possible explanation is
that inter-firm effects are not allowed for in the CBI data. Company
expectations concerning employment were much closer to the actual
history of employment, however. Mayer suggested that this was because
firms effectively control their own employment levels and therefore are
reporting their plans rather than their expectations.
Mayer speculated that the poor performance of the expectations model for
orders and sales might be due to the fact that forecasting procedures
vary across firms. This had lead Mayer and Mors to develop a model that
allowed the expectations data to be generated by a mixture of different
forecasting methods, and this approach proved much more successful.
Timo Terasvirta (University of Helsinki) suggested that the
expectations data were difficult to model because they reflected the
views of a variety of firms, each subject to different production lags.
Did the results indicate that firms are pessimistic in their forecasts,
Terasvirta wondered? Mayer explained that the method of converting the
qualitative expectations data into quantitative form had assumed that on
average expectations were correct. He also speculated that the results
showed that firms place too much weight on the theoretical model and too
little on the observed data.
The workshop stimulated lively interchanges among the participants, and
it is hoped to follow up soon with further collaborative activities,
initially in applications of game theory and disequilibrium
macroeconomics and econometrics.
|
|