RDP 7704: Money and Money–Income: An Essay on the “Transmission Mechanism” II. The Transmission Mechanism in an IS-LM Context

The basic IS-LM model is well-known. Where E is real private ex-penditure, A is autonomous real private expenditure, Y is real income, G is real government expenditure, M the quantity of nominal money and P the general price level, its static form can be written, in linear terms for simplicity.

This model, whose expenditure function is sometimes augmented by a wealth effect, and which, for empirical purposes, is inevitably dynamized by introducing distributed lags into the structural equations that make it up, is concerned with the determination of aggregate demand. If we adopt the standard textbook “reverse L” shaped aggregate supply curve, and consider the horizontal section of it along which the price level is constant, we may derive the following reduced form equation for income

Clearly, in terms of this model, questions about the relationship between money and income concern the second term on the right-hand side of this reduced form, and these questions may be divided into two groups. First the appropriateness of treating the behavior of the quantity of money as determining that of money income has often been disputed. this is the “reverse causation” question. Secondly, there are questions about the stability and size of the money multiplier considered in isolation, and compared with the autonomous expenditure multiplier.

The reverse causation question need not detain us long at this stage though we will have a good deal to say about it below. Let it simply be noted that it has long been widely recognized that the monetary authorities can act in such a way as to ensure that variations in the quantity of money are the effect rather than the cause of variations in income—for example, by gearing their open market operations to pegging the rate of interest at a particular value. It is also indisputable that there have been historical episodes in which this has been done explicitly, for example, in the United States over the period 1941–51. Questions about reverse causation have usually been treated as being, by their very nature, questions about the conduct of monetary policy during particular historical episodes.

Questions about the money multiplier seem, from the point of view of the IS-LM research agenda, to be of a fundamentally different nature. They concern behavioral relationships in the private sector of the economy whose characteristics are to be regarded as independent of the conduct of policy. Thus quantitative knowledge about the money multiplier is to be thought of as conferring not only the power to answer questions about the channels whereby a given monetary policy did influence economic activity during a particular episode, but also to answer questions about how alternative monetary policies would have influenced activity had they instead been implemented. Moreover, such knowledge is also to be viewed as enabling predictions about the consequences of future policies to be made so that their design may be improved. As we shall see below, much recent work questions the proposition that the structure of the economy is independent of the nature of the policy being carried out, but that proposition is a fundamental premise of the research on the transmission mechanism of monetary policy which is now to be discussed.

It is obvious from equation (3) that if the value of the money multiplier is to remain stable over time, then so must the parameters of both the demand for money function and the aggregate expenditure function. It is equally obvious, not to say well-known, that its value, both absolute and relative to the autonomous expenditure multiplier, is particularly sensitive to the values of the parameters a and ℓ which measure the sensitivity of expenditure and demand for money.^[3] Thus, it is hardly surprising that the role of interest rates in influencing agents behavior has been at the center of empirical research concerned with the transmission mechanisms of monetary policy. Laidler (1977) extensively surveys both theoretical and empirical work on the demand for money function. Suffice it here to say that the existence of a stable aggregate demand for money function is well established by empirical work, and in particular, that a clearly defined negative relationship between the demand for money and an interest rate variable is a salient characteristic of that function. There is no question of the extreme “quantity theory” special case of the IS-LM model being supported by empirical evidence; this fact implies that, if we take that model seriously, the nature of the relationships underlying the parameter a, both qualitative and quantitative, is of central importance to the linkage between money and economic activity.

The aggregate private sector expenditure function embodied in equation (1) is a convenient simplification. Expenditure on currently produced goods and services by the private sector involves the behavior of both firms and households. Firms expenditure may be on producer durable goods as well as on inventories of raw materials and finished output. Households, on the other hand, buy both nondurable and durable goods. Even so, the qualitative nature of the linkage between money and expenditure which the IS-LM model attempts to summarize has not been a subject of much substantive controversy, although some have questioned the adequacy of the summary that it gives of that linkage.^[4] Money is commonly regarded as one of a spectrum of assets held by firms and households, whose (not always very clearly specified) services yield diminishing marginal utility (or product) to their consumer. An increase in the quantity of money in an economy initially in asset equilibrium thus induces a disequilibrium in the structure of asset holding because the implicit yield on money is thereby driven down. A generalized substitution from money into other assets takes place, driving down their rates of return. Some of these rates of return are observable interest rates on securities, set by specialist dealers in response to supply and demand conditions in organized markets, some are observable borrowing and lending rates set by financial intermediaries of one sort or another, and others are implicit, nonobservable, rates of return on assets such as consumer durables. Such a general fall in rates of return involves an increase in the present value of the income streams yielded by existing assets, and a rise in their market values above the supply price of newly produced assets. The output of durables, both consumer and producer, therefore increases. At the same time, the price of current consumption in terms of future consumption foregone has fallen so that in principle an increase in expenditure on nondurables might also be expected to occur.^[5] Whatever the category of expenditure upon which impact effects fall, they are amplified by a multiplier process which is only partially offset by the subsequent behavior of interest rates.

The argument of the previous paragraph says nothing about the quantitative significance of the various effects discussed, and it is here that substantive disagreements have arisen.^[6] Nevertheless, the qualitative issues raised do have important implications for the interpretation of empirical evidence on the transmission mechanism, particularly that evidence which comes from certain large-scale econometric models. In principle such models are capable of opening up to inspection the interior of the “black box” that connects monetary policy to the behavior of money income. In practice, these models, with certain notable exceptions (the Canadian RDX2 model, for example, and to a lesser degree the U.S. FMP model) have frequently omitted monetary variables from all but a subset of expenditure functions—typically involving firms investment decisions—and in thus narrowing down the channels of causation that they investigate virtually ensure that only relatively weak links between money and economic activity will be discovered. This is not to say that the results these models produce are necessarily wrong, but it is to say that what should be empirical questions have all too often been settled a priori in constructing them.^[7]

Because it is possible to construct a theoretically coherent case that all aspects of expenditure might respond to monetary variables, some would argue that it is necessary to investigate all aspects of expenditure empirically before one can conclude that monetary factors are, or are not, important. Others take the argument further. If reverse causation can be ruled out, and if a significant correlation between money and money income exists, then this, in and of itself, is prima facie evidence that a transmission mechanism between the variables also exists. This is the reasoning that underlies the preference of many monetarists for testing for the importance of monetary factors in the economy by way of reduced form equations. If one accepts it, then detailed work on the various aspects of the transmission mechanism should be interpreted not as. investigating the existence of such a mechanism, but as seeking information about its nature.^[8]

In fact, of course, a variety of approaches have been taken to investi-gating the questions about linkages between monetary variables and economic activity that the IS-LM prompts. Reduced form equations, of the type fitted to U.S. data by Friedman and Meiseiman (1963) and Andersen and Jordan (l968) certainly show that there is a positive and statistically significant re-lationship between the time paths of the quantity of money and money income, both over long runs of data and in the post “accord” period as well. The relationship between money and income also dominates that between autonomous expenditure and income. Moreover, as far as more recent data are concerned, the statistical analysis of Sims (1972) strongly suggests that causation runs predominantly from money to money income rather than vice versa.^[9] For earlier periods Friedman and Schwartz (1963) (1970) using the approach of the historian rather than the econometrician conclude that causation has run in both directions, with that running from money to income predominating.

Reduced form studies have been carried out for Britain by Barrett and Walters (1966) using data for the period 1878–1963, and by Artis and Nobay (1969) as well as the Bank of England (1970) for post-Second World War data. Here too there can be no doubt that correlations between money and money income exist, but it is a fair generalization that these are not nearly so strong and well determined as their counterparts for the United States. It is also notable that when Williams, Goodhart, and Gowland (1976) applied Sims techniques to British data for the period 1958–1971, no clear-cut results on the direction of causation between money and money income emerged. If anything, this study supports the “reverse causation” hypothesis. As to earlier periods, Howson's (1975) work on interwar Britain is, like that of Friedman and Schwartz, based on historical rather than econometric methods. Although she attributed an important causative role to monetary variables at certain times, notably during the upswing that followed the abandonment of gold in 1931, she found it hard to attribute much influence to money in other episodes, for example, in determining the economy's cyclical behavior in the mid- and late 1920s. Problems like those encountered with British evidence occur in the Canadian case as well. Macesitch's (1966) (1969) attempts at replication of the Friedman and Meiselman study for that economy led to a debate which left the significant of the money-income relationship, and the direction of causation, for that economy, an open question.^[10] Furthermore, Barth and Bennett's (1974) replication of Sims test for Canada for the years 1957–1972 showed, at best (and then only provided that money was narrowly defined), that the interaction between money and money income involved causation running in both directions, and hence did nothing to solve the problems left open by the earlier work.

It is certainly easier to find a clear-cut correlation between money and money income for the United States than for Britain or Canada, but the details of an IS-LM type of transmission mechanism are just as hard to pin down for that country as for others. For the United States, the whole question of the importance of monetary factors in influencing business investment in fixed plant and equipment is currently an open one. Jorgenson's theoretical work (e.g., 1967) on the neoclassical theory of investment clarified the role played by the rate of interest in determining the oppor-tunity cost of the services of capital equipment, and the empirical work which he and his associates (e.g., Jorgenson and Stevenson (1967), Jorgenson, Hunter and Nadiri (1970)) carried out seems to show that investment responds with a time lag, to interest rate variations. However, that work has been criticized by Brechling (1974) (1975) on a number of grounds.

The time lags in Jorgenson's empirical work were introduced and specified arbitrarily. Brechling shows that the neoclassical theory of investment supplemented by an adjustment coat hypothesis, itself has definite implications, for the time pattern of investment's response to interest rates, that ought to be incorporated specifically into the empirical formulation of time lags. He also shows that the neoclassical theory of investment, either as specified by Jorgenson, or with the addition of lags arising from adjustment costs, is not robust in the face of empirical evidence. The particular structural equation derived from that theory, that Jorgenson and his associates chose to fit to data, seems to perform well enough, but that equation treats output as an exogenous variable. The reduced form expression implied by the same theory which has prices, wages and the cost of capital (in which an interest rate variable is incorporated) as exogenous variables, but not output, fits badly with wrongly signed parameters being the rule rather than the exception. Brechling also shows that the equation that Jorgenson and his associates fit is not an appropriate reduced form for a cost-minimization, as opposed to profit-maximization, formulation of the neoclassical theory and that the appropriate reduced form fits the data badly. Thus, though interest rates might well systematically influence business investment by way of neoclassical mechanisms, Brechling's work shows that this proposition remains to be demonstrated as far as the United States economy is concerned.^[11] It also, incidentally, cast doubts upon the account of the transmission mechanism implied by the FMP econometric model, since that model's treatment of investment draws heavily on Jorgenson's work.

One must be careful not to infer from the foregoing analysis that there is no evidence that business investment is affected by monetary factors. As Lund (1971) and fisher and Sheppard (1974) note, there is abandoned evidence, from studies based on one form or another of the accelerator hypothesis, that output or sales variables seem to be important determinants of investment. Jorgenson's own empirical work, of course, leads to similar results. If monetary factors affect consumer expenditures, then sales or output variations, to the extent that these are the result of monetary changes, provide an important indirect channel whereby those same changes influence investment.^[12] This point is worth stressing, because, for the United States at least, there is a good deal of evidence to show consumer expenditure is sensitive to monetary influences. Thus Hamburger (1967) has found that interest rates played an important role in determining the demand for new durable goods over the period 1953–1964. The demand for newly constructed housing is also well-known to be sensitive to monetary factors, with credit availability effects whose roots lie in imperfections in the mortgage market playing a significant role, at least according to the FMP model. That same model takes a “life cycle” approach to formulating the consumption function, and the effects of interest rate variations on the value of stocks owned by households exert an influence on their levels of expenditure that constitutes an important component of the transmission mechanism which the model generates.

For Britain and Canada the evidence on all these issues is at least as mixed as it is for the United States. For both countries there do exist studies that find a significant role for interest rates in determining invest-ment—Hines and Catephores (1970) or the investment equations of the London Business School (LBS) model for Britain, and the investment equations of both the RDX2 and CANDIDE 1.2 models of Canada, to give some examples. With the exception of Hines and Catephores, all this work draws heavily and explicitly on that of Jorgenson, but none of it has been subjected to the same thorough scrutiny which Brechling brought to bear on studies of United States data.

Accelerator type effects do seem to be important for investment in both countries, and, for Britain, Trivedi (1970) has found a marginally significant role for interest rates to play in determining inventory investment, a result which seems to have eluded those working with U.S. and Canadian data.

The influence of monetary variables on consumption is less well established for Britain and Canada than for the U.S. In particular, the type of wealth effects that figure so prominently in the FMP model seem to be barely present in British data (cf. Deaton (1972)). The influence of Interest rates as relative prices on the demand for consumer durables is recognized in the Canadian RDX2 model while they are also allowed to influence overall savings behavior. In this model, also, credit availability influences the housing market though, in CANDIDE, the monetary influences are trans-mitted to this market by interest rates. There is little debate about the importance of availability effects on that market as far as Britain is con-cerned. Even when the “Radcliffe view” about the relative unimportance of monetary policy was at the height of its popularity, the availability of mortgage funds was regarded as a key factor in the new housing market, and events of recent years have done nothing to alter anyone's views on this issue.^[13] It is hardly surprising, given the frequency with which legal restrictions on down-payments and the periods of consumer loans were varied in the 1950s and 1960s that credit availability effects seem to be important in influencing the demand for consumer durables in Britain (cf. Hilton and Crossfield (1970) and Garganas (1975)).

Now the evidence sketched out here would, taken at face value, suggest that “money matters” to a greater extent in the United States economy than in Britain, and perhaps than in Canada also. Crude correlations between money and money incomes are better determined in United States data; there seems less ambiguity about the direction of causation between the variables in those same data; and the details of an IS-LM type of transmission mechanism are at least as easy to discern in the case of the United States. However, an alternative, and simpler, interpretation of the evidence is that the IS-LM framework, although reasonably well adapted to the study of the United States, particularly the United States of the 1950s and 1960s from which so many of the results we have cited above are derived, is, in general, an inadequate model for investigating the links between money and money income.

Three well-known shortcomings of the IS-LM approach to macroeconomic questions are particularly relevant to this point of view. First, the model does not deal satisfactorily with the fact that variations in money income are made up of fluctuations in real income and prices; secondly, it ignores potential linkages between the government's budget and the behavior of the money supply; and thirdly, it is a model of a closed economy. It seemed until recently that these problems merely required that the model be extended, that equations be added to it. Indeed, the large-scale econometric models mentioned above all do extend the framework in one way or another to cope with these factors. It was not apparent that such extensions would also require us to modify our views about the qualitative nature of the linkages between money and aggregate demand that we have already discussed. However, as we shall now see, recent theoretical work has indeed forced us to do just that, and hence casts doubt upon the extent to which the evidence cited so far enables us to come to grips with a number of key questions about the transmission mechanism.

Footnotes

It is easy to recast this argument concerning their importance for the money and autonomous expenditure multipliers in terms of elasticities instead of slope parameters. Simply note that while , and multiply by . I owe this point to Michael Parkin. [3]

It is, for example, a basic proposition of the work of Brunner and Meltzer (e.g., 1976) that the markets for money, credit, and real capital ought to be explicitly present in any analytic framework, and that, in neglecting credit markets, the IS-LM model is crucially deficient. [4]

This account of the transmission mechanism is widely accepted. It appears essentially as presented here in the work of Tobin (see 1969) , Friedman (see Friednan and Meiselman, 1963), and Brunner and Meltzer (see Brunner and Meltzer, e.g., 1976). Note that Brunner and Meltzer in particular stress the behavior of the price of currently existing physical assets relative to the supply price of new assets as a basic element in the transmission mechanism. As already noted, they also stress that, once assets other than money and physical capital are admitted explicitly into the analysis it becomes necessary to consider the elasticities of demand and supply for such assets with respect to the rate of interest as well as the interest elas-ticities of demand for money and investment goods. Whether this makes a qualitative difference or merely a quantitative difference to one outcome of any analysis depends, as Dornbusch (1976) has shown, on whether the sharpest break in the chain of substitution between money and physical capital occurs at the liquid end or the less liquid end of the spectrum of assets. [5]

The extreme Keynesian view that only the interest sensitivity of investment is relevant to the transmission mechanism has not been found commonly in recent North American literature. However, some British econo-mists still seem to regard this as the critical step in the mechanism. [6]

The foregoing paragraph draws heavily upon Fisher and sheppard (1974). Among the models that rely heavily-perhaps too heavily-on interest rates' influence on investment to link the monetary and real sectors are the 1972 version of the London Business School model of the United Kingdom and the Economic Council of Canada CANDIDE 1.2 model. See Ball et al. (1975) and Economic Council of Canada (1975) respectively. For an account of the links between money and economic activity in the FMP model, see Modigliani (1975) and for RDX2, see Helliwell et al. (1971). [7]

In Laidler (1971), 1 criticized reduced form studies. This criticism was not directed at the use of reduced forms per se, but rather at the dif-ficulties that arise in interpreting them if they are not related to an explicitly formulated model. Neither Friedman and Meiselman (1967) nor Andersen and Jordan (1968) derived their equations from an explicitly formu-lated model. [8]

Sims' test, building on the work of Granger (1969), essentially relies upon the timing of variations in the data. If, after the relevant series have been prefiltered to remove serial correlation from regression residuals, variations in income are more closely correlated with preceding variations in the quantity of money, than variations in the quantity of money are with preceding variations in money income; if variations in money income are less closely correlated with succeeding variations in the quantity of money, than variations in the quantity of money are with succeeding vari-ations in money income; then it is argued that the predominant direction of causation runs from money to money income. Note that Fisher and Sparks have argued (1975) that it is only possible to use data on timing to establish directions of causation in the context of a specific theoretical framework. Note also that Feige and Pearce (1976b) argue that Sims' results are ex-tremely sensitive to variations in the prefiltering procedures used. [9]

For an account of the debate as far as Canadian data are concerned see Fisher and Sparks (1975) . [10]

Brechling suggests that the supply side as well as the demand side of the capital goods market needs to be analyzed explicitly and brought into the picture before satisfactory tests concerning the interest sensitivity of investment demand can be carried out. In this his work echoes the stress that Brunner and Meltzer lay on the role played by the supply price of newly-produced capital assets in the transmission mechanism. [11]

Note that this is not a new argument. It was explicitly set out as long ago as (1939) by A. J. Brown. [12]

The mechanism involved here involves the borrowing and lending rates of institutions involved in the mortgage market being sticky. Then a rise in market interest rates elsewhere attracts funds from these institutions, leading to a shortage of mortgages, at going interest rates, and hence to credit rationing in the housing market. The mechanism is described in the Radcliffe Report (1959), paragraphs 292–295. Arcelus and Meltzer (1973) report no such effects to be present in the U.S. mortgage market. [13]