I. Introduction

On the basis of the most recent flow-of-funds accounts, Italian households’ net wealth is estimated to have reached about 2,500 trillion lire by the end of 1985, having closely kept in line with the evolution of households’ disposable income (Table 1). ^1/ Indications are that, as in other European countries, a large proportion of total wealth is held by households in the form of real assets (dwellings, agricultural land, and consumer durables), although the share of dwellings in real wealth tends to be greater in Italy than in other countries. With respect to other European countries, however, the financial liabilities of Italian households represent a relatively small component, totaling only about 6 percent of gross financial assets and showing a tendency to decrease over time.

Table 1.

Composition of Italian Households’ Net Wealth

(In percent)

Sources: Banca d’Italia (1985a, 1987).Note: “Other assets” include banker’s acceptances, atypical securities, Bank of Italy cashier’s checks and checks. “Other liabilities” include bad debts on loans of banks, insurance companies, and social security institutions and, from 1984 onward, nonbank consumer credit.

Table 1.

Composition of Italian Households’ Net Wealth

(In percent)

				1975	1980	1985	1986
Real assets				72.1	73.9	65.4	…
	Of which:
		Dwellings		81.0	80.8	83.2
		Land		9.8	9.2	5.5
		Durables		9.2	9.9	11.3
Financial assets				30.4	28.1	36.7	…
	Of which:
		Notes and coins		6.8	5.4	3.8	3.4
		Bank deposits		44.5	45.2	34.1	30.0
		Postal deposits		8.2	8.0	5.6	5.5
		Treasury bills		0.1	8.6	12.7	10.7
		Deposits and savings
			certificates of special
			credit institutions	2.6	1.6	2.3	2.1
		Treasury credit certificates		—	2.3	12.0	11.4
		Other government securities		3.1	2.3	2.1	3.3
		Other securities		8.2	2.7	2.7	2.5
		Italian shares and participations		5.3	7.0	9.5	13.5
		Foreign assets		1.4	1.0	1.0	1.1
		Actuarial reserves and severance pay provision		19.8	15.5	11.6	16.5
		Other assets		0.1	0.5.	2.4
Financial liabilities				2.6	2.0	2.2	…
	Of which:
		Bank loans		39.9	56.4	53.5	56.6
		Special credit institutions loans		56.5	37.8	32.4	31.2
		Other liabilities		3.7	5.9	14.1	12.2
Net wealth/disposable income				4.4	4.6	4.4	…

Sources: Banca d’Italia (1985a, 1987).Note: “Other assets” include banker’s acceptances, atypical securities, Bank of Italy cashier’s checks and checks. “Other liabilities” include bad debts on loans of banks, insurance companies, and social security institutions and, from 1984 onward, nonbank consumer credit.

View Table

Table 1.

Composition of Italian Households’ Net Wealth

(In percent)

				1975	1980	1985	1986
Real assets				72.1	73.9	65.4	…
	Of which:
		Dwellings		81.0	80.8	83.2
		Land		9.8	9.2	5.5
		Durables		9.2	9.9	11.3
Financial assets				30.4	28.1	36.7	…
	Of which:
		Notes and coins		6.8	5.4	3.8	3.4
		Bank deposits		44.5	45.2	34.1	30.0
		Postal deposits		8.2	8.0	5.6	5.5
		Treasury bills		0.1	8.6	12.7	10.7
		Deposits and savings
			certificates of special
			credit institutions	2.6	1.6	2.3	2.1
		Treasury credit certificates		—	2.3	12.0	11.4
		Other government securities		3.1	2.3	2.1	3.3
		Other securities		8.2	2.7	2.7	2.5
		Italian shares and participations		5.3	7.0	9.5	13.5
		Foreign assets		1.4	1.0	1.0	1.1
		Actuarial reserves and severance pay provision		19.8	15.5	11.6	16.5
		Other assets		0.1	0.5.	2.4
Financial liabilities				2.6	2.0	2.2	…
	Of which:
		Bank loans		39.9	56.4	53.5	56.6
		Special credit institutions loans		56.5	37.8	32.4	31.2
		Other liabilities		3.7	5.9	14.1	12.2
Net wealth/disposable income				4.4	4.6	4.4	…

Sources: Banca d’Italia (1985a, 1987).Note: “Other assets” include banker’s acceptances, atypical securities, Bank of Italy cashier’s checks and checks. “Other liabilities” include bad debts on loans of banks, insurance companies, and social security institutions and, from 1984 onward, nonbank consumer credit.

Since the mid-1970s, households’ net wealth allocation appears to have been characterized by two major trends. First, as far as the composition of total wealth is concerned, the real component showed a sizable decline toward the mid-1980s, coinciding with a sharp drop in the relative value of dwellings. Second, among financial assets, there was a substantial shift from M2 money (currency in circulation, plus bank and postal deposits) to government securities. The latter increased from nearly 3 percent of financial wealth in 1975 to more than 13 percent in 1980 and to more than 25 percent in 1986. At the same time, the M2 share declined by nearly 20 percentage points, to 39 percent in 1986. The substitution away from currency and deposits took place almost entirely in the first half of the 1980s, since in the late 1970s most of the increase in the share of government securities was at the expense of nongovernment medium-term and long-term securities whose share more than halved, from about 8 percent in 1975 to less than 3 percent in 1980.

To put the whole discussion in the right perspective, notice that developments in the allocation of financial wealth reflect most of the features that have characterized and, to a lesser extent, still characterize the Italian financial structure. The crucial point is that, historically, Italian sectoral financial balances have consistently shown a pattern characterized by households’ financial surpluses approximately balancing the public sector’s increasing deficits. The corresponding need for a high degree of financial intermediation, coupled with the limited development of the Italian capital market, has paved the way for a predominant role for credit institutions in financial intermediation. The high and variable rates of inflation of the mid-1970s and the consequent collapse of the bond market further strengthened the phenomenon of overintermediation by the banking system, leading to the absorption of the economy’s financial assets into bank deposits.

It was only in the second half of the 1970s that the explosion in the public sector borrowing requirement, the outburst of inflation (and the consequent inability of the bond market to finance the deficit), and the evolution of monetary policy (Caranza and Fazio, 1983) fostered the unprecedented growth of short-term and floating rate securities such as treasury bills (Buoni Ordinari del Tesoro, BOT) and treasury credit certificates (Certificati di Credito del Tesoro, CCT) ^2/ that started to be accumulated in households portfolios. Subsequently, the continuous growth of public debt and monetary policies aimed at encouraging its absorption by the household sector (as reflected in the positive ex post real interest rates prevailing since 1982) further encouraged financial investment, particularly in government securities.

Developments such as these raise a number of important issues. First, it is legitimate to ask whether the “effective” stock of money should also include, besides deposits and currency, other financial instruments (with appropriate weights) such as short-term government securities. Second, continuing debt-financed government deficits raise the question of whether (and to what extent) government debt drives up the rate of return on capital, thus crowding out interest-sensitive private sector spending (in particular, residential investment and investment in new productive capacity). Third, the Italian inflationary experience calls for an assessment of the responsiveness of the demand for government securities to the inflation rate, because this impinges on the complete bond refinanceability of inflation-induced interest service. This, in turn, is related to the issue of the appropriate measurement of fiscal deficits.

In addition, it is often suggested that relative price movements cannot be invoked as the only explanatory factor for such dramatic shifts in the composition of households’ wealth. Shifts in monetary aggregates “unaccounted for” by interest rate behavior are frequently pointed out and attributed to the role played by financial innovations, in particular (i) as long as they lower transactions costs for purchases and sales of assets, and/or (ii) they lead to the appearance of previously unavailable assets. ^3/ Undoubtedly, such innovations have taken place extensively in the Italian financial markets in recent years, and the recent popularity of mutual funds—repackaging assets to yield different services—provides the most recent such example. However, merely recognizing the existence of specific innovations is only a starting point: how they also affect agents’ behavior in quantitative terms remains to be understood.

From an empirical viewpoint, the available econometric evidence in Italy is still scanty. The only available references on the allocation of (financial) wealth are Modigliani and Cotula (1973), Cotula and others (1984), and Banca d’Italia (1986b). In all cases, however, the whole private nonbank sector is considered. In view of this, the main purpose of the present paper is to provide additional empirical evidence regarding household demand for monetary assets.

The traditional way of attacking the problem, following the standard discrete-time theory of risk-averse portfolio choice and under well-known assumptions, would be to regress actual portfolio shares held by households against some measure of expected returns, exploiting (more or less completely) the information derived from portfolio optimization theory. As a result, a number of important questions that depend on the paramaters of investors’ demands for assets could be tentatively answered (see Friedman (1978, 1985), Roley (1982), and Frenkel (1985), among others).

The status of this kind of empirical exercise is, however, not entirely undisputed. Friedman (1985) frankly reports the discouraging results of applying maximum likelihood methods to estimate investors’ behavior using data on the aggregate portfolio, of U.S. households. Frankel and Dickens (1983) test the hypothesis of mean-variance optimization and statistically reject it. The poor empirical performance of traditional portfolio models is not an unexpected result, given the central importance of time-varying investors’ risk perceptions in governing how market-clearing expected returns respond to changes in supplies of assets, and considering also the restrictive representation of aggregate preferences used in applied work.

A possible alternative would be to infer investors’ unobservable perceptions of asset risks from actual after-tax return data, and then to apply the standard theory of risk-averse portfolio allocation to recover the relevant parameters of investors’ demands for assets. However, even abstracting from the issue of measurement of expected rates, ^4/ this line of research appears to be largely unfruitful in the Italian case because of the lack of an adequate data base. In fact, data for long-term holdings do not contain market valuation changes, quarterly data for equity holdings are not available, and the evaluation of tangible assets such as real estate and consumer durables is still somewhat doubtful.

In the light of these comments, the paper focuses on households’ demand for short-term assets in a context of “money in the utility function.” Along with some recent literature (Diewert (1974), Donovan (1978), Barnett (1980, 1981), Ewis and Fisher (1984, 1985), Husted and Rush (1984), Serletis and Robb (1986), and Poterba and Rotemberg (1986)), the paper assumes the households’ objective function to be defined over nondurable consumer goods, labor services and the services of consumer durables, and monetary assets. Given standard (but not unquestionable, intertemporal and intratemporal) separability assumptions, derived demand functions for financial assets are specified and estimated. The important advantages of working with explicit portfolio models involving utility maximization include testable within-equation and cross-equation restrictions corresponding to a meaningful representation of preferences, which can be used to improve the overall efficiency of estimation. Furthermore, it is quite natural to incorporate into this framework recent advances in the literature on functional forms, thereby introducing more flexibility into the analysis.

As is well known, this approach “is a reduced-form approach which models, restricts, and characterizes the results of the consumer’s decisions without the need to consider the explicit structure of the decision. While the ‘true’ utility function does not contain money, a derived utility function containing money generally can be acquired” (Barnett (1981), p. 191). However, it is a controversial approach. Its critics dislike the idea that money yields utility directly, and they point out that assets that are rate-of-return-dominated are held because they reduce transaction costs, which should be explicitly modeled. The implicit modeling upon which the utility approach is based, however, is not necessarily alternative to the explicit modeling used in the transaction-demand approach. Indeed, recent research (Feenstra (1986)) has shown that, under specific assumptions, a functional equivalence exists between treating real balances as an argument of the utility function and entering money into liquidity costs as a part of the budget constraint.

Admittedly, some of the applied literature mentioned above can be criticized in principle on the ground that the consumer’s portfolio allocation problem tends to be modeled as one of choosing expenditures (i.e., opportunity cost times the quantity held) on liquidity services, and hence on different assets, while the opportunity cost of these assets is a random variable at the time that the consumer allocates his portfolio. This difficulty is avoided in the recent work of Poterba and Rotemberg (1986): building up on Hansen and Singleton (1982), they estimate the parameters of a representative consumer’s utility function from, the first-order condition of the maximization problem. But this line of research presents a number of serious problems. Its critics (Deaton (1986), among others) have underlined the danger of its application to aggregate data. Furthermore, its empirical implementation is limited to specific functional forms that unduly restrict preferences, and which have therefore been dismissed by empirical demand analysis a long time ago. ^5/ In the light of these comments, this paper follows standard practice and ignores the uncertainty of returns and assets.

In order to focus on the details of demand for services on monetary assets, the paper ignores the upper (intertemporal) stage of the consumer’s optimization problem, that is, the allocation of the consumer’s resources among broad categories (consumption goods, labor, and monetary services). In this framework, four main issues are addressed. First, we specify and present estimates of the parameters of the technology allowing households to combine financial instruments available in the market, and we derive a nonmarketed good such as “liquidity services.” These estimates imply a whole pattern of substitution among financial assets and allow a study of the effects of changes in interest rates on holdings of assets.

Second, we allow short-term and floating rate government securities such as BOTs and CCTs to provide liquidity services and present an indirect test of this contention. The inclusion of government securities in the liquidity-generating technology is by no means unusual (Barth, Kraft, and Kraft (1977), Ewis and Fisher (1984, 1985), and Poterba and Rotemberg (1986)), and it appears to be suitable (at least as a working hypothesis) in the Italian case, where both short-term and medium-term debts (indexed to the short-term interest rate) are widely regarded as being easily marketable. In this respect, a more traditional risk-return framework should not necessarily be preferred but could instead be usefully confined to the choice between long-term debt, equities (together amounting to some 15 percent of households’ financial wealth), and tangible assets.

Third, as in Barnett (1980) and Barth, Kraft, and Kraft (1977), we exploit the concept of “committed quantities” and model the dynamic process governing the system of equations on demand for assets with a general procedure consistent with the “habit-formation” hypothesis.

Finally, assuming that financial innovations can be classified as affecting one or more of the relevant coefficients (Judd and Scadding (1982)), but not necessarily the preference parameters, we adopt a flexible functional form and model the elasticity of substitution between assets as a time-varying quantity. At the same time, we test a number of interesting restrictions on preferences, and we attempt to detect the presence of structural changes in the coefficients of the monetary aggregator implicit in the technology-generating demand for assets.

The paper is structured as follows. Section II introduces the theoretical framework and discusses a specific functional form for the households’ objective functions. Section III presents the estimation procedure and the empirical results. Section IV uses the estimated parameters to provide empirical evidence on the structure of the households’ demand for financial assets. Section V concludes the paper and outlines some policy implications.

II. Theoretical Background

Drawing on Diewert’s (1974) seminal paper as well as on the subsequent work of Donovan (1978) and Barnett (1978, 1980, 1981), some of the recent literature on the demand for assets has incorporated many of the concepts and techniques of modern consumer theory. Exploiting recent advances in the duality theory, systems of asset-demand equations have been derived from household models of utility-maximizing behavior by applying the concept of the rental price of durable goods to financial assets.

Formally, given the appropriate separability assumptions (Barnett (1981)), the optimization problem of the representative household can be cast in terms of the following one-period minimization:

Define:

where m_t = (m_1t, …, m_It) is an I-vector of quantities of monetary assets (i.e., stocks deflated by some general price index, p_t) held in period t and assumed to generate liquidity services proportionately; their rental prices (Jorgensonian user costs) are given by the I-vector v_t = (v_1t, …, v_It). Hence, M_t represents total expenditure on liquidity services, while C(•) is a cost function defining the minimum cost of reaching a given level of utility, u, at a given price vector, v, and is assumed to satisfy the usual regularity conditions.

For the i-th asset,

which denotes the discounted interest forgone by holding a lira’s worth of that asset. In equation (3), R_t is the expected one-period holding (including realized or unrealized capital gains or losses) nominal yield available on a benchmark asset, that is, an asset accumulated to transfer wealth between multiperiod planning horizons, rather than to yield liquidity services during the current period. Notice that, from the agent’s viewpoint, R_t contains all the relevant premiums available in the market for forgoing the services provided by monetary assets. Furthermore, r_it is the expected one-period holding nominal yield of the i-th asset. ^6/

It is worth noting that in equation (3) nominal interest rates can be expected to respond to expected inflation rates. In addition, since Jorgensonian user costs for consumer durables depend inversely upon the expected inflation rate, it follows that expected inflation can influence demand for assets through a number of interesting channels.

Despite its theoretical nicety, the empirical application of this approach must face the problem of giving an appropriate definition to the benchmark asset. The available literature has usually singled out a particular financial asset such as equity holdings (as in Poterba and Rotemberg (1986)), or long-term corporate bonds (as in Husted and Rush (1984) and Donovan (1978)), or, in each period, the asset yielding the highest rate of return (as in Serletis and Robb (1986) and Ewis and Fisher (1984, 1985)). In the present paper we identify the asset as long-term corporate bonds and tangible assets (dwellings).

To provide a sufficient degree of flexibility, we consider, for equation (1), the following functional form:

where

with

Bollino (1987) shows that equations (4)-(6) generalize Deaton and Muellbauer’s (1980) Almost Ideal (A.I.) demand system, by introducing overhead costs or committed quantities (that is, μ_i’s) into the original cost function belonging to the Piglog class. In terms of the primal problem, equation (4) implies that utility is generated by the net quantities (m_i - μ_i). Alternatively, equation (4) can be understood as the counterpart of a budget constraint defined in terms of expenditure net of committed expenditure consumed independently of current rates of return or expenditure on monetary services. In other words, the agent must bear a fixed (overhead) cost before obtaining any utility from liquidity services.

The μ_i provides an immediate and theoretically coherent, well-known way of introducing dynamics into the system. Since Pollak (1970), the importance of past decisions as determinants of present allocation patterns has been recognized, in order to characterize both short-run and long-run agent preferences. In the present context we adopt a general specification (known as “dynamic translating” since Pollak and Wales (1981), and already present in Barnett (1980)), which permits some preference parameters to vary with past choices and allows a “habit-formation” interpretation. In particular, we define

The “dynamic translating” procedure is general in that it is not functional form-dependent and can be applied to any demand system that represents the static maintained hypothesis. ^7/

Reverting to equations (4)-(6), by Shephard’s lemma the following share equations can be obtained:

where

Equations (8) and (9) define a system of demand equations adding up to total expenditure on liquidity services, homogeneous of degree zero in rental prices and total expenditure, and symmetric in the substitution matrix if the following restrictions are satisfied:

Unfortunately, there is no way of imposing, by means of simple parametric restrictions, the negative semidefiniteness of the substitution matrix.

Following Deaton and Muellbauer’s (1980) suggestion, in order to simplify the estimation procedure, log V will be approximated as follows:

on the grounds that if rental rates move together, then log V will be quite insensitive to the choice of weights.

III. Estimation and Empirical Results

In the empirical application we consider quarterly Italian time series ^8/ of five categories of financial assets for the third quarter of 1977 through the fourth quarter of 1986. ^9/ The five aggregates are (1) “notes and coins and demand deposits;” (2) “savings deposits;” (3) “postal deposits and savings and credit certificates of special credit institutions;” (4) “treasury bills;” and (5) “treasury credit certificates.”

A number of points regarding the treatment of the data should be made. First, since 1984, Italian investors have been given the opportunity to invest even small amounts in a block, of generically defined securities managed by investment funds. Presumably, the substantial increase in investment fund resources led to a corresponding reduction of the direct demand for securities by savers, crowding out in particular the direct demand for CCTs, which by the end of 1986 amounted to about 40 percent of mutual funds’ asset holdings. Since the amount of information available is certainly insufficient to allow separate treatment of investment fund units, it was decided to allocate this component of household financial wealth pro quota to CCTs. While this procedure is at best a rough approximation, it substantially improves the forecasting performance of the model (as we should expect under reasonable assumptions), while leaving other features of the estimates unaltered.

Second, in the light of the previous discussion, R is defined as a weighted average of the one-period holding nominal yield available on long-term corporate bonds and the rate of return on real assets, with weights given by the actual weights of those assets in total wealth normalized to add up to unity.

Third, with the exception of long-term assets, all rates of return are defined as actual, net of taxes, one-period holding nominal yields. Capital gains and losses are tentatively taken into account for long-term assets, computing the returns as in Banca d’Italia (1986b). ^10/

Fourth, a premium accounting for the difficulties encountered in evaluating the rate of return on real assets is estimated by scanning and is added to R. It turns out to equal 6 percent on a yearly basis.

The parameters of the system (equations (7), (8), and (10)), estimated by Full Information Maximum Likelihood, are presented in Table 2 along with some summary statistics. ^11/ Most of the coefficients in Table 2 are significant based on individual tests, and the overall fit seems satisfactory. Three of the β_i’s are significantly different from zero, thereby pointing to a prima facie violation of the commonly accepted hypothesis of a liquidity-generating technology independent of scale. ^12/ Nine out of ten free γ_ij’s are also significantly different from zero and they testify to the substantial price responsiveness of agents. As we would expect (given their relationship with adjustment speeds), the λ_i’s are all significantly different from zero and are of an increasing magnitude as we move from short- to longer-term assets. The Durbin-Watson statistics, although not entirely appropriate in the present context, do not suggest dynamic misspecification of the system.

Table 2.

Parameter Estimates Under Homogeneity and Symmetry Constraints (Third Quarter 1977 - Fourth Quarter 1986)

Notes:i = 1: notes and coins and demand deposits;i = 2: savings deposits;i = 3: postal deposits and savings and credit certificates ofspecial credit institutions;i = 4: treasury bills;i = 5: treasury credit certificates.

Table 2.

Parameter Estimates Under Homogeneity and Symmetry Constraints (Third Quarter 1977 - Fourth Quarter 1986)

	i = 1	i = 2	i = 3	i = 4	i = 5
μi/100	-.417	-.883	-.085	-.029	-.015
	(.202)	(.436)	(.048)	(.098)	(.037)
λi	.280	.622	.514	.651	.921
	(.131)	(.136)	(.101)	(.054)	(.048)
αi	.397	.625	-.036	.239	-.224
	(.073)	(.079)	(.040)	(.070)	(.071)
βi	-.028	-.020	.019	-.028	.057
	(.016)	(.018)	(.009)	(.015)	(.020)
γii	.201
	(.016)
γ2i	-.115	.162
	(.026)	(.027)
γi3	-.042	-.029	.077
	(.008)	(.007)	(.012)
γi4	-.038	-.016	-.001	.047
	(.011)	(.010)	(.003)	(.017)
γi5	-.007	-.002	-.005	.008	-.006
	(.004)	(.004)	(.002)	(.003)
${\hat{\mathrm{\sigma }}}^{\mathrm{2}}\mathrm{\text{x100}}$	.264	.305	.121	.277	.190
dw	1.73	2.16	1.42	2.16	2.25

Notes:i = 1: notes and coins and demand deposits;i = 2: savings deposits;i = 3: postal deposits and savings and credit certificates ofspecial credit institutions;i = 4: treasury bills;i = 5: treasury credit certificates.

View Table

Table 2.

Parameter Estimates Under Homogeneity and Symmetry Constraints (Third Quarter 1977 - Fourth Quarter 1986)

	i = 1	i = 2	i = 3	i = 4	i = 5
μi/100	-.417	-.883	-.085	-.029	-.015
	(.202)	(.436)	(.048)	(.098)	(.037)
λi	.280	.622	.514	.651	.921
	(.131)	(.136)	(.101)	(.054)	(.048)
αi	.397	.625	-.036	.239	-.224
	(.073)	(.079)	(.040)	(.070)	(.071)
βi	-.028	-.020	.019	-.028	.057
	(.016)	(.018)	(.009)	(.015)	(.020)
γii	.201
	(.016)
γ2i	-.115	.162
	(.026)	(.027)
γi3	-.042	-.029	.077
	(.008)	(.007)	(.012)
γi4	-.038	-.016	-.001	.047
	(.011)	(.010)	(.003)	(.017)
γi5	-.007	-.002	-.005	.008	-.006
	(.004)	(.004)	(.002)	(.003)
${\hat{\mathrm{\sigma }}}^{\mathrm{2}}\mathrm{\text{x100}}$	.264	.305	.121	.277	.190
dw	1.73	2.16	1.42	2.16	2.25

Notes:i = 1: notes and coins and demand deposits;i = 2: savings deposits;i = 3: postal deposits and savings and credit certificates ofspecial credit institutions;i = 4: treasury bills;i = 5: treasury credit certificates.

Tables 3, 4, and 5 report a sequence of tests of theoretical interest. In Table 3, ψ identifies the usual likelihood ratio test and r the number of constraints. In addition, when appropriate, we also report the degrees of freedom-corrected likelihood ratio statistics ψ. (Pudney (1981), p. 575) and ψ₂ (Meisner (1979), p. 232). Apart from checking whether the data are consistent with the assumption of utility-maximizing behavior, the hypothesis of no overhead costs, of quasi-homotheticity, and of restricted price response are also statistically tested. The results are of considerable interest in that some of the hypotheses that have usually been assumed to characterize systems of demand for assets tend to be rejected by the data. Homogeneity and symmetry cannot be rejected on the basis of the available evidence. The assumptions of zero-committed quantities and of no-habit formation are strongly at variance with the data while the standard assumption of quasi-homotheticity ^13/ is only mildly so. The same conclusion can be drawn from the test for restricted price responses, which is obtained by letting all γ_{i j}’s equal zero, thereby reducing the A.I. system of demand to a slightly dnderparameterized member of the same family belonging to the Forgeaud Nataf (1959) class, in which demands are a function of real total expenditure and the relative own price alone. As Deaton (1976) has shown, this would imply an approximate proportionality between expenditure and price elasticities, which is usually regarded as rather implausible behavior in applied demand analysis. On the contrary, the data suggest that flexibility is a basic ingredient in modeling demand for assets price responses and these restrictions are soundly rejected.

Table 3.

Likelihood Ratio Tests

Table 3.

Likelihood Ratio Tests

		r	ψ	ψ1	ψ2
Homogeneity
	[H0: Σj γij = 0 (∀i)]	4	11.6	5.5	7.5
Symmetry
	[H0: γij = γji(∀i, j)]	6	7.6	—	—
Overhead cost
	[H0: μi = λi = 0 (∀i)]	10	263.4	249.9	187.3
Constant overhead
	[H0: μi = 0 (∀i)]	5	43.2	36.3	30.7
Habit formation
	[H0:λi = 0 (∀i)]	5	148.6	141.7	105.7
Quasi-homotheticity
	[H0: βi = 0 (∀i)]	4	14.6	9.1	10.4
Restricted price response
	[H0: γij = 0 (∀i, j)]	10	73.0	59.5	51.9

View Table

Table 3.

Likelihood Ratio Tests

		r	ψ	ψ1	ψ2
Homogeneity
	[H0: Σj γij = 0 (∀i)]	4	11.6	5.5	7.5
Symmetry
	[H0: γij = γji(∀i, j)]	6	7.6	—	—
Overhead cost
	[H0: μi = λi = 0 (∀i)]	10	263.4	249.9	187.3
Constant overhead
	[H0: μi = 0 (∀i)]	5	43.2	36.3	30.7
Habit formation
	[H0:λi = 0 (∀i)]	5	148.6	141.7	105.7
Quasi-homotheticity
	[H0: βi = 0 (∀i)]	4	14.6	9.1	10.4
Restricted price response
	[H0: γij = 0 (∀i, j)]	10	73.0	59.5	51.9

Table 4.

Weak Separability Test

Table 4.

Weak Separability Test

		r	ψ
Reference year: 1978
	[H0: u(ul(M2), u2(BOT, CCT))]	6	7.0
	[Ho: u(u1(M3), u2(CCT))]	6	17.1
Reference year: 1980
	[H0: u(ul(M2), u2(BOT, CCT))]	6	6.3
	[H0: u(u1(M3), u2(CCT))]	6	19.6
Reference year: 1982
	[H0: u(ul(M2), u2(BOT, CCT))]	6	5.7
	[H0: u(ul(M3), u2(CCT))]	6	27.0
Reference year: 1984
	[H0: u(u1(M2), u2(B0T, CCT))]	6	5.6
	[H0: u(ul(M3), u2(CCT))]	6	21.1
Reference year: 1986
	[H0: u(ul(M2), u2(BOT, CCT))]	6	7.3
	[H0: u(u1(M3), u2(CCT))]	6	17.1

View Table

Table 4.

Weak Separability Test

		r	ψ
Reference year: 1978
	[H0: u(ul(M2), u2(BOT, CCT))]	6	7.0
	[Ho: u(u1(M3), u2(CCT))]	6	17.1
Reference year: 1980
	[H0: u(ul(M2), u2(BOT, CCT))]	6	6.3
	[H0: u(u1(M3), u2(CCT))]	6	19.6
Reference year: 1982
	[H0: u(ul(M2), u2(BOT, CCT))]	6	5.7
	[H0: u(ul(M3), u2(CCT))]	6	27.0
Reference year: 1984
	[H0: u(u1(M2), u2(B0T, CCT))]	6	5.6
	[H0: u(ul(M3), u2(CCT))]	6	21.1
Reference year: 1986
	[H0: u(ul(M2), u2(BOT, CCT))]	6	7.3
	[H0: u(u1(M3), u2(CCT))]	6	17.1

Table 5.

Parameter Stability Test

Note: Overall sample: Third Quarter 1977 - Fourth Quarter 1986. Section I denotes stability tests performed including mutual funds’ holdings of government bonds into CCTs, with Section II excluding them.

Table 5.

Parameter Stability Test

		r	ψ	ψ2
Estimation excluding:
I. First year		16	21.6	15.3
	First two years	32	29.1	20.7
	Last two years	32	70.1	49.8
	Last year	16	27.5	19.6
II. First year		16	21.9	15.6
	First two years	32	30.9	22.0
	Last two years	32	81.0	57.6
	Last year	16	42.7	30.4

Note: Overall sample: Third Quarter 1977 - Fourth Quarter 1986. Section I denotes stability tests performed including mutual funds’ holdings of government bonds into CCTs, with Section II excluding them.

View Table

Table 5.

Parameter Stability Test

		r	ψ	ψ2
Estimation excluding:
I. First year		16	21.6	15.3
	First two years	32	29.1	20.7
	Last two years	32	70.1	49.8
	Last year	16	27.5	19.6
II. First year		16	21.9	15.6
	First two years	32	30.9	22.0
	Last two years	32	81.0	57.6
	Last year	16	42.7	30.4

Note: Overall sample: Third Quarter 1977 - Fourth Quarter 1986. Section I denotes stability tests performed including mutual funds’ holdings of government bonds into CCTs, with Section II excluding them.

Table 4 presents an indirect test of the hypothesis of weak separability of a money aggregate with respect to a government securities aggregate, in the uncommitted technology. ^14/ The test, which is a Wald test basically testing the implications of the Leontief-Sono condition, is conditional on a given portfolio allocation and, therefore, was computed for alternate years after 1978. As it turns out, throughout the period the marginal rate of substitution between the components of M2 money does not seem to depend on short-term and floating rate government securities and, therefore, an M2 aggregate can be meaningfully defined. Given the nature of CCTs, however, an M3 aggregate (i.e., M2 plus treasury bills) is never appropriately defined.

As an additional check on the performance of the model, Table 5 reports the result of estimating the model excluding the observations for one or two years at the beginning or at the end of the sample period and then computing the structural stability test proposed in Anderson and Mizon (1983). In all cases, the hypothesis of a stable model structure cannot be rejected at conventional significance levels if proper account is taken of the tendency of these tests to reject the null in small samples. Interestingly, though, parameter stability tends to fall by the end of the period if mutual funds’ holdings of government bonds (mostly CCTs) are not merged with households’ holdings of the same asset.

In conclusion, these results tend to reinforce the contention that casting the analysis in terms of more general, flexible models may capture some of the complexities of the available empirical evidence. In particular, it seems that during the period under study, financial innovations have not necessarily implied structural breaks, that is, changes in the parameter characterizing the underlying representation of consumer preferences.

IV. The Structure of Italian Households’ Demand for Monetary Assets

Having estimated and tested the model, we can now use its parameters to study the composition of households’ portfolios and the effects of changes in interest rates on asset holdings. It should be noted that in the present case analytical solutions for the long-run preferences are not available. Therefore, it proved necessary to revert to dynamic simulations. These were performed, first, by computing the long-run portfolio structure based on end-1986 levels of total short-term financial wealth and interest rates. Subsequently, each rate of return was shocked by one percentage point in order to trace and assess the pattern of price responses for a given level of total short-term financial wealth. ^15/

In considering the long-run, fully-adjusted portfolio structure implied by end-1986 rate differentials, it is interesting to note that the share of postal deposits and BOTs remains approximately unchanged. In comparison, the share of CCTs over total assets is substantially larger (plus 6 percentage points, or, equivalently, 50,000 billion lire), at the expense of bank deposits. In other words, there is reason to believe that the substitution away from bank deposits would continue, given the present rates.

Since the theoretical framework adopted in this paper allows for time-varying elasticities, Table 6 presents price responses evaluated at end-1986 variable levels. For simplicity and comparability, price responses are presented in terms of the two most relevant aggregates, that is, M2 money and short-term and floating rate government securities.

Table 6.

Effects of One-Point Change in Selected Rates of Return

(Reference year: 1986, percentage points)

Note: Price responses are computed for a given level of total short-term financial wealth.

Table 6.

Effects of One-Point Change in Selected Rates of Return

(Reference year: 1986, percentage points)

		Quarter	Money	BOT and CCT
Rates of return:
	Demand and savings deposits	0	0.80	-0.97
		4	2.19	-2.62
		LR	2.98	-3.54
	Demand deposits	0	0.17	-0.20
		4	0.27	-0.32
		LR	0.29	-0.35
	Savings deposits	0	0.60	-.73
		4	1.88	-2.25
		LR	2.71	-3.22
	Postal deposits	0	-0.04	0.05
		4	-0.34	0.51
		LR	-1.03	1.23
	BOTs and CCTs	0	-1.53	1.84
		4	-3.98	4.75
		LR	-4.97	5.91
	CCTs	0	-1.17	1.41
		4	-4.69	5.61
		LR	-8.19	9.74

Note: Price responses are computed for a given level of total short-term financial wealth.

View Table

Table 6.

Effects of One-Point Change in Selected Rates of Return

(Reference year: 1986, percentage points)

		Quarter	Money	BOT and CCT
Rates of return:
	Demand and savings deposits	0	0.80	-0.97
		4	2.19	-2.62
		LR	2.98	-3.54
	Demand deposits	0	0.17	-0.20
		4	0.27	-0.32
		LR	0.29	-0.35
	Savings deposits	0	0.60	-.73
		4	1.88	-2.25
		LR	2.71	-3.22
	Postal deposits	0	-0.04	0.05
		4	-0.34	0.51
		LR	-1.03	1.23
	BOTs and CCTs	0	-1.53	1.84
		4	-3.98	4.75
		LR	-4.97	5.91
	CCTs	0	-1.17	1.41
		4	-4.69	5.61
		LR	-8.19	9.74

Note: Price responses are computed for a given level of total short-term financial wealth.

Price responses in Table 6 should be understood as the percentage change in the demand for the relevant financial aggregate following a one-point change in the rate of return of the j-th asset. Responses are recorded in the same quarter in which the change in the rate of return takes place (quarter 0), one year after (quarter 4), and in the long run, or five years after (LR).

Comparing these with the corresponding responses evaluated at different points in time (not reported here owing to space limitations), it is possible to appreciate the advantage of working with models flexible enough to accommodate time-varying responses while keeping the underlying structure of preferences constant.

To start with, in 1986 a one-point change in the rates of return on (both demand and savings) bank deposits would have led in equilibrium to a 3 percent increase in M2 money (corresponding to about 13,000 billion lire), and a 3.5 percent decrease in short-term government securities. As expected, agents would shift from short-term government securities to deposits in order to “produce” liquidity services efficiently. It is worth noting that performing the same calculations for 1980 implies quite different price responses, reflecting a different portfolio structure. M2 money would have increased by less than 2 percent, with short-term and floating rate government securities decreasing by more than 7 percent.

This result hides important differences in the price responsiveness of different financial instruments. In particular, if the rates of return on demand and savings deposits are increased separately, an entirely different picture emerges. In fact, changes in the former rate only imply a reallocation in M2 money: demand deposits rise by nearly 3 percent in equilibrium and savings deposits fall by more than 4 percent, with the demand for government securities decreasing only marginally. In comparison, changes in the latter rate lead to a reshuffling of the entire portfolio: BOTs and CCTs decrease by more than 2 percent and 3 percent, respectively; demand deposits fall by almost 1 percent, while savings deposits rise by more than 6 percent.

An even stronger pattern of interactions is implied by a one-point change in the rate of return on short-term government securities. Recalling that the CCT rate is linked to the BOT rate, such a change would lead in equilibrium to a 6 percent increase in the stock of these securities (corresponding to nearly 20,000 billion lire) and a slightly less than 5 percent decrease in M2 money. Again, the reference period is crucial in evaluating these responses. In 1980, a one-point change in the rate of return on BOTs and CCTs would have produced in the long run a meager 2.5 percent increase in the demand for bank and postal deposits, while, in the light of the then limited market dimension, the demand for BOTs and CCTs would have grown by nearly 10 percent.

Once more these figures hide asset-specific effects. Consider, for example, a 1 percent increase in the CCT spread over the BOT rate. This would lead, with respect to the previous case, to an even higher increase in the overall demand for government securities (nearly 10 percent in equilibrium) resulting, however, from a massive shift of all assets into CCTs. The demand for CCTs would increase by more than 15 percent after one year and 20 percent in equilibrium, mainly at the expense of BOTs (minus 10 percent after one year and minus 14 percent in equilibrium) and postal deposits (minus 6 percent and minus 10 percent, respectively).

Unfortunately, not all results based on the estimated parameters lend themselves to meaningful interpretation. In particular, changes in the rate of return on postal deposits appear to produce quantitatively negligible, albeit qualitatively perverse, responses that deserve further investigation.

Finally, another interesting exercise, motivated by the nonhomotheticity of the system, is given by a 1 percent increase in total short-term assets for given levels of the rental rates. In the short run, bank deposits absorb a relatively higher proportion of a shock in the level of total assets, as clearly implied by impact elasticities of about 1.3 for M2 money and 0.6 for government securities. One year later, the situation is entirely reversed: the scale elasticity of M2 money drops to 0.8 (reaching 0.5 in equilibrium), while the scale elasticity of government securities increases to 1.3 (1.7 in equilibrium).

V. Inflation and the Demand for Government Securities

An interesting application of the model estimated in Section III results from the assessment of asset demand responses to changes in the expected inflation rate, and from its implications for the adjustment for inflation of conventional measures of government deficits.

A common argument is that the depreciation of the real value of government bonds owing to inflation should be subtracted from debt service charges when computing the deficit, in order to achieve a better understanding of the potential impact of the deficit. This argument assumes full bond refinanceability of inflation-induced interest services and, therefore, tends to ignore the potentially relevant portfolio effects of inflation. In other words, it is assumed that the government is able to issue and place, without altering returns and liquidity conditions, new securities to finance the depreciation component of debt service charges.

If, however, inflation induces asset substitution—for example, government securities for Ml money—then it is crucial to recognize that an increase in inflation could prove expansionary even though the adjusted deficit remains constant.

In the model estimated in the present paper, portfolio effects of inflation can occur in two distinct ways. On the one hand, changes in expected inflation should, in principle, affect only the opportunity cost of holding demand deposits and cash, thus leading to traditional substitution effects. On the other hand, they can influence the overall demand for liquidity services and, given the nonhomotheticity of the system, the share of government securities.

As far as the latter channel is concerned, estimation of the overall demand for liquidity services is beyond the scope of this paper. Therefore, we revert to the former channel and notice that in Table 2 the estimated parameters imply sizable substitution effects.

Computing elasticities using 1986 as the reference year, the pattern of substitution is characterized chiefly by a switch from demand deposits and cash to savings deposits and, to a lesser extent, to BOTs and CCTs, as depicted in Table 7. A similar picture emerges when computing elasticities in the early 1980s.

Table 7.

Effects of One-Point Change in Expected Inflation Rate

(Reference year: 1986, percentage points)

Note: Responses are computed for a given level of total short-term financial wealth.

Table 7.

Effects of One-Point Change in Expected Inflation Rate

(Reference year: 1986, percentage points)

		Quarter
		0	4	LR
Assets:
	Demand deposits and cash	-0.28	-0.46	-0.46
	Savings deposits	0.42	0.78	0.80
	Postal deposits	-0.29	-0.60	-0.62
	BOTs	0.05	0.07	0.06
	CCTs	0,02	0.03	0.03

Note: Responses are computed for a given level of total short-term financial wealth.

View Table

Table 7.

Effects of One-Point Change in Expected Inflation Rate

(Reference year: 1986, percentage points)

		Quarter
		0	4	LR
Assets:
	Demand deposits and cash	-0.28	-0.46	-0.46
	Savings deposits	0.42	0.78	0.80
	Postal deposits	-0.29	-0.60	-0.62
	BOTs	0.05	0.07	0.06
	CCTs	0,02	0.03	0.03

Note: Responses are computed for a given level of total short-term financial wealth.

In summary, the impact of expected inflation on government securities is limited in magnitude. Abstracting from other effects of inflation, it can be concluded that these figures do not constitute any substantial evidence against adjusting the government deficit for inflation.

VI. Concluding Comments

In this paper we have specified, estimated, and tested a utility-function approach to the demand for monetary assets and short-term and floating rate government securities using Italian quarterly data for the household sector for 1977-86. Although much work remains to be done, our results suggest a number of conclusions of both theoretical and empirical interest.

First, from an econometric modeling point of view, the paper has clarified the importance of working with flexible functional forms when estimating systems of equations of demand for assets that are characterized, as in the Italian case, by dramatic compositional changes. Rather than invoking structural breaks to compensate for functional failures, this approach can interpret substantial data variation while keeping the underlying structure of preferences constant.

Second, from a policy point of view, the empirical results suggest that bank deposits, in generating liquidity, show a sizable degree of substitution with respect to short-term and floating rate government securities. Indeed, the present (end-1986) level of total financial wealth and the structure of rate of return differentials imply in the long run a substitution away from bank deposits, in particular into CCTs, far beyond what is already taking place.

Third, the degree of “moneyness” of short-term and floating rate government debt, although not negligible, does not endanger the meaning of traditional monetary aggregates such as M2 money. At the same time, however, the pattern of substitution among financial instruments allows both short-term and floating rate government securities to be lumped together in a single aggregate.

Fourth, it appears that inflation induces asset substitution to a very limited extent and leads to a fall in the demand for M2 money and a marginal rise in the demand for government securities. The portfolio effects of inflation of the kind considered in this paper are not such as to deter adjustment of government deficit for inflation.

Finally and more generally, the evidence points to a high degree of responsiveness by households to relative price signals—a conclusion with important implications, considering the increasing integration of financial markets both inside and outside the country.