BANK OF GREECE

Economic Bulletin

Number 14, December 1999

BANK OF GREECE 21, E. Venizelos Avenue 102 50 Athens Economic Research Department - Secretariat Tel. (030-1) 320 2392 Fax (030-1) 323 3025 Printed in Athens, Greece at the Bank of Greece Printing Works ISSN 1105 - 9729

BANK OF GREECE

Economic Bulletin

Number 14, December 1999

6

ECONOMIC BULLETIN, 14 12/99

Contents

The real exchange rate and economic activity: is the hard-drachma policy necessarily contractionary?

7

Leading inflation indicators for Greece: an evaluation of their predictive power

19

Estimation and analysis of external debt in the private sector

43

Monetary policy and banking supervision measures

55

Bank of Greece decisions

59

Statistical section

61

The real exchange rate and economic activity: is the hard-drachma policy necessarily contractionary?1

1. Introduction The relationship between the real exchange rate and the level of economic activity is a very important, as well as controversial, issue. This is due to the fact that the real exchange rate affects gross domestic product (GDP) through various transmission mechanisms, which operate through both the demand for, and the supply of, total output. The traditional literature mentions that, in case of depreciation of the real exchange rate, aggregate demand will increase, as the gain in international price competitiveness of domestic goods improves the current account balance. This presupposes that the sum of

Christos E. Papazoglou Economic Research Department

price elasticities of export and import demand is higher than unity.2 On the other hand, and in contrast to the traditional view, there are theoretical approaches, according to which a real depreciation can lead to economic contraction. First of all, a nominal depreciation can affect adversely economic activity, if the initial expansionary effect, achieved through the traditional shift in expenditure from imported to domestic goods, is fully offset by the reduction in aggregate demand caused by the rise in prices (Alexander 1952). Moreover, real depreciation limits aggregate demand, through its effect on the distribution of income. Specifically, according to Krugman and Taylor (1978), a real depreciation can lead to income redistribution from individuals with a low marginal propensity to save to those with a high marginal propensity, thus causing a decrease in demand and, consequently, in output.

1 The article expresses the author’s opinions and not necessarily those of the Bank of Greece. I wish to thank G. Zombanakis, N. Zonzilos, P. Momferatos, D. Moschos, S. Pantazidis and G. Hondroyianis for their very useful comments. 2 The traditional literature refers, for the most part, to the original Keynesian approach. See e.g. Alexander (1959), Tsiang (1961) and Machlup (1966).

ECONOMIC BULLETIN, 14 12/99

7

Beyond the effect it has through aggregate demand,

as we have mentioned, focuses mostly on the neg-

a real depreciation exerts considerable influence on

ative effect a real appreciation has on aggregate

the level of economic activity through aggregate

demand. Consequently, it largely ignores possible

supply effects. For example, it can reduce aggregate

positive effects stemming from the other transmis-

supply, since it leads to an increase in the cost of

sion mechanisms mentioned previously. For exam-

imported raw materials, thereby reducing their im-

ple, it usually overlooks the positive effect of the

ports and, consequently, the level of domestic pro-

reduced cost of imported raw materials on aggre-

duction (Hanson 1983). Argy and Salop (1983), as

gate supply and competitiveness, especially in the

well as Lizondo and Montiel (1989), claim that the

case of small economies with a strong dependence

fall in profits from non-tradeable goods, which is

on imports of raw materials. Chart 1 depicts

caused by the higher cost, due to the depreciation,

Greece’s real GDP and the weighted real ex-

of imported raw materials (especially oil, the most

change rate of the drachma (based on the Con-

important raw material for most countries), leads to

sumer Price Index). Given that a rise in the exchange

a drop in production.

rate is equivalent to a currency appreciation, it follows that these two variables show a strong posi-

Empirical studies of the impact of a real deprecia-

tive correlation from 1987 onwards.

tion on total product usually focus on developing countries. For example, Branson (1986), Edwards

This study examines empirically the effect of

(1986) and Rhodd (1993) have examined the con-

changes in the real exchange rate of the drachma on

sequences of a devaluation for the GDP of sev-

the production level of the Greek economy. More

eral developing countries. More recent studies by

specifically, we attempt to estimate a reduced form

Halpern and Wyplosz (1996), as well as Mitchell

equation for real gross domestic product, derived

and Pentecost (1998), tackle the same subject for

from structural equations capturing both the aggre-

the transition economies of Central and Eastern

gate demand for, as well as the supply of, output.

Europe.3

The structure of the model allows us to incorporate in the GDP equation explanatory variables such as

The relationship between the real exchange rate

the real exchange rate, the nominal wage, the quan-

and production is an important and, almost always,

tity of money and public expenditure.

a timely issue for the Greek economy. Specifically, for a long period until the drachma’s devaluation in

The results of the econometric analysis suggest that

March 1998, it was claimed that the constant

the appreciation of the drachma’s real exchange

appreciation of the drachma’s real exchange rate

rate causes only a temporary decline in real GDP,

led directly to the deterioration of the current

which recovers subsequently, as the economy

account balance, mainly through the reduction in exports. It therefore had a negative effect on the level of economic activity. Indeed, the policy of slow depreciation of the drachma relative to inflation was usually called the “hard-drachma policy”. This view adopts the traditional approach which,

8

3 The results of these studies are often contradictory. For example, Branson’s (1986) study of the Kenyan case shows that the devaluation of the local currency has an important contractionary effect on the economy. On the other hand, the studies by Edwards (1986) for 12 developing countries and by Rhodd (1993) for Jamaica show that this contractionary effect is temporary and that real GDP increases in the long term.

ECONOMIC BULLETIN, 14 12/99

The real exchange rate and economic activity

moves towards long-term equilibrium. The chief

supply, in which prices are determined by a mark-

conclusion is that the hard-drachma policy fol-

up on costs, while the production level is deter-

lowed in recent years in Greece as the main tool of

mined by aggregate demand. It is an extension of

the anti-inflationary effort does not necessarily have

the Dornbusch (1976) model and is closer to

a contractionary effect on economic activity.

Sachs (1980) and Marston (1985) models, which also include the supply side. Concerning the effect

The study proceeds as follows: the next section pre-

of changes in the real exchange rate on the level

sents a simple structural macroeconomic model,

of production, we rely on the following two hypo-

which is used for the derivation of the real GDP

theses: the first, which concerns the aggregate

equation. The model is an augmented form of the

demand side, assumes that a real appreciation has

well-known IS-LM system, since it includes the sup-

a contractionary effect on output, owing to the

ply side of the economy. The third section explains

deterioration of the current account balance caused

the econometric methodology and the empirical

by the worsening of competitiveness. Therefore,

results that stem from the estimation of the model.

on the basis of the previous analysis, the effect

Finally, the last section presents some conclusions.

through the traditional transmission mechanism is assumed to prevail over the other factors that affect output through aggregate demand. The

2. The model

second hypothesis refers to the aggregate supply side and expresses the expansionary effect of

As mentioned above, the analysis relies on a

the appreciation on output through the reduction

simple model of aggregate demand and aggregate

in production costs. Hence, the relative strength of

ECONOMIC BULLETIN, 14 12/99

9

these two effects determines how the apprecia-

where ¢ = (‚ + ·‰) > 0 and q ≡ p–e–p*.

tion of the real exchange rate affects real GDP.

Therefore, aggregate demand for domestic output, which incorporates money market equilibrium as

We begin the presentation of the model from the

well, depends directly on the real quantity of

aggregate demand side. Specifically, the demand

money, expected inflation, the real exchange rate

for output is defined as:

and the level of real government expenditure.

y = – ‰(r – ) – Û(p – e – p*) + g

(1)

The aggregate supply side is based on a simple production function of the form:

where y is the logarithm of real total product, r the nominal interest rate,  the expected rate of infla-

Y = ÊN

(4)

tion, g the logarithm of real government expenditure, and p–e–p* the real exchange rate, where e

According to equation (4), the level of production

is the logarithm of the price of the foreign cur-

depends directly on the labour input N. Produc-

rency expressed in the local currency, while p and

tion units employ workers up to the point where

p* are the logarithms of the domestic and the for-

the marginal product of labour, Ê, equals the real

eign level of prices, respectively. We note that the

wage (W/P). This approach allows us to express

real exchange rate has been defined in such a way

prices, P, as a percentage of the nominal wage, W.

that its rise corresponds to a real appreciation of

Specifically,

the domestic currency. According to equation (1), demand for real domestic product is negatively

P = [(1 + Ì)/Ê] W

(5)

affected by both the real interest rate and the real where Ì is the constant profit margin.4 If we take

exchange rate.

into account the real or relative price of raw mateEquilibrium in the money market is given by the

rials, Pm / P, equation (5) is expanded as follows:

equation: P = [(1 + Ì)/Ê] W (Pm / P)ı m – p = ·y – ‚r

(6)

(2) where coefficient ı represents the inflow of raw

where m is the logarithm of the quantity of

materials per unit of output and, therefore, (Pm/P)ı

money. The coefficients · and ‚ represent the elasticity and semi-elasticity of money demand with respect to income and the interest rate, respectively. Solving equation (2) for r, which is then replaced in equation (1), we have: y = ¢–1[‰(m – p) + ‚‰ – ‚Ûq + ‚g]

10

(3)

4 The profit margin may be affected by changes in the real exchange rate, especially under monopoly conditions. See e.g. Blanchard and Muet (1993). This argument is reinforced by the Greek experience in the period following the devaluation of the drachma in March 1998. Specifically, the prices set by industrial and commercial firms, with very few exceptions, rose by more than was justified by the increase in the cost of imports associated with the devaluation (see Bank of Greece 1998, 1999). We should note, however, that the constant profit margin hypothesis does not alter the results of the analysis in terms of quality.

ECONOMIC BULLETIN, 14 12/99

The real exchange rate and economic activity

represents the part of the per-unit costs which is

the quantity of money causes a fall in interest rates

due to imported raw materials.

and thus contributes to an increase in output through the rise in aggregate demand. Real government expenditure has a direct positive effect on the

The logarithmic form of equation (6) is

level of output. On the other hand, the rise in the p = k + w – ıq

(7)

nominal wage raises production costs and leads to an output fall. Finally, the result of a real apprecia-

where k = log[(1 + Ì)/Ê], logW = w and logPm =

tion is ambiguous, because of two counterbalancing

(e + p*). As we see, according to (7), raw materi-

effects which stem from aggregate demand and sup-

als are purchased at international prices.

ply, respectively. Initially, as equation (10) shows, the appreciation of the real exchange rate leads to a

Finally, concerning the formation of inflationary

direct and an indirect decline in aggregate demand.

expectations, we adopt the simple hypothesis that

The direct effect is due to the deterioration in the

the price level approaches the level of equilibrium –, with a constant speed of adjustment, Á. prices, p

international competitiveness of domestic products,

Specifically,

The indirect effect manifests itself through the

and the main consequence is a drop in net exports. impact of the real appreciation on the real interest

– – p), where 0 ≤ Á ≤ ∞.  = Á( p

(8)

rate. More specifically, a rise in the real exchange rate limits inflationary expectations, leading to a re-

Therefore, inflation rises when p– > p, drops when p– < p and remains stable when p– = p. The level

duction in demand for capital goods because of the

of equilibrium prices is determined by the inter-

hand, the rise in the real exchange rate leads to an

national level of prices, denominated in the domestic currency, that is, p– = (e + p*), which means

expansion of output supply by lowering the costs of

that equation (8) can be expressed as follows:

costs. Hence, on the basis of the theoretical model,

resulting rise in the real interest rate. On the other

imported raw materials and, therefore, production the effect of the appreciation of the real exchange

 = Á(e + p* – p) = – Áq

(9)

rate on domestic output is ambiguous.5

If we substitute equations (7) and (9) into the aggregate demand equation, that is, equation (3),

3. Empirical methodology and results

the equation for output becomes: The next stage of the analysis concerns the empiry = ¢–1{‰k + ‰m – ‰w – [‚(Û + ‰Á) – ‰ı]q + ‚g} (10)

ical estimation of equation (10) for the Greek

Given that ¢ > 0, we expect the following signs: ∂y / ∂m > 0, ∂y / ∂w < 0, ∂y / ∂g > 0 and ∂y / ∂q ≥≤ 0. According to the above partial derivatives, a rise in

ECONOMIC BULLETIN, 14 12/99

5 We note that the model does not directly examine the role of international capital movements. Certainly, to the degree that capital movements affect either the exchange rate (in the case of floating rates) or the quantity of money (in the case of fixed exchange rates), their effect on output is included in the values of the respective parameters of these variables in equation (10).

11

economy. The general logarithmic form of the model

real exchange rate, which is based on the con-

to be estimated is as follows:

sumer price index. Moreover, we examine whether use of the other indices affects the results of the

y = b0 + b1m + b2w + b3q + b4g + ut

(11)

analysis.

The above equation is used in the empirical inves-

The econometric methodology is based on Engle

tigation of the contribution of the drachma’s real

and Granger’s (1987) two-stage method. We ini-

exchange rate to changes in the real GDP of

tially tested for the non-stationarity of the vari-

Greece, after taking account of changes in the

ables on the basis of the augmented Dickey-Fuller

quantity of money, the nominal wage and real

(ADF) tests and the relevant results are presented

government expenditure.

in Table 1.7 Next, we tested for the existence of a long-run relationship between the variables, that

Equation (11) was estimated on the basis of quar-

is, we tested for the existence of cointegration.

terly data on the Greek economy from the first

This was done by testing the hypothesis of non-

quarter of 1980 through the fourth quarter of 1997.

stationarity of the residuals in equation (11).

The variables used were: real GDP for y, the M3 nominal quantity of money for m, the nominal

The results of the estimation of equation (11) are

hourly wage for w, the real effective exchange rate

presented in Table 2.a. According to the ADF

of the drachma, deflated by the consumer price

tests, we reject the hypothesis that residuals are

index for q, and real government expenditure for

non-stationary; this fact suggests the existence of

g. The data were obtained from Bank of Greece’s

a cointegration relationship among the variables

publications Monthly Statistical Bulletin and

in equation (11), which expresses the long-term

Statistical Bulletin of Economic Conjuncture, the

relationship among these variables.8 As Table 2.a

International Monetary Fund’s International

shows, all independent variables have the expected

Financial Statistics, and the National Statistical

signs. Concerning the real exchange rate, the results

Service of Greece’s (NSSG) National Accounts.6

show that an appreciation of the drachma has a

Concerning the derivation of the real exchange rate, it should be noted that, besides the CPI, we used the wholesale price index and unit labour costs. Since each of the above indices has certain disadvantages, their interpretation needs special care. Specifically, unit labour costs do not include other production cost components, the wholesale price index excludes services, and the consumer price index includes indirect taxes, thereby distorting the price competitiveness of exported goods and services. As mentioned above, the empirical analysis that follows uses the definition of the

12

6 It should be noted that the quarterly data on real GDP, for the period up to 1988, have been estimated by the NSSG, while those for the period 1989-1997 are estimates by the Economic Research Department of the Bank of Greece. 7 According to the results of this test, all the variables are stationary in first difference; their levels are non-stationary. 8 We note that the long-term relationship, as it appears in equation (11), does not express the notion of a steady state. Specifically, as is known from theory, the long-run equilibrium suggests that the real variables of the economy are not affected by changes in nominal variables, while the real exchange rate is determined by the international purchasing power parity condition (see e.g. Dornbusch 1976). In the present analysis, the longrun effect of the appreciation of the real exchange rate on output refers to the medium term, which follows the initial (short-run) effect of the appreciation, before the economy reaches its longrun equilibrium.

ECONOMIC BULLETIN, 14 12/99

The real exchange rate and economic activity

Table 1 Unit root tests Variable

ADF statistics y q m W g

Variable

ADF statistics

0.48 0.2 8.6 13.9 0.35

¢y ¢q ¢m ¢w ¢g

-9.95** -11.77** -7.16** -3.25** -12.9**

Note: ADF statistics follow the t distribution and are used to test the hypothesis that the respective time series is I(1). ADF critical values at significance levels of 1 per cent: –2.59 and 5 per cent: -1.95. ** Indicates a significance level of 1 per cent.

After estimating the long-term relationship, the

Table 2 Empirical results

short-term relationship linking the above variables can be estimated using Granger’s representation

2.a Co-integration relationship Variables

theorem. Specifically, the short-term relationship Dependent variable y

c q m w g

1.5 0.45 0.4 -0.37 0.17

Residuals ADF statistics

-8.1**

is presented as an error-correction model, which incorporates, besides the variables’ first differences, another term which includes the stationary residuals of the long-term relationship with one period time lag, which show the deviation from equilibrium. The usefulness of the error-correction

2.b Error correction model Variables

model is derived from the fact that most times the Dependent variable ¢y

¢y(-4) ¢q(-4) u(-1)

2

Adjusted R Standard error

economy is not in long-term equilibrium. It is usu-

0.35* (2.6) -0.23* (-2.0) -0.52* (-2.34)

ally in a non-equilibrium state and requires time

0.95 0.024

will fall if it exceeds its equilibrium level, while the

Note: The figures in parentheses are t statistics. The test of the autocorrelation hypothesis is performed with the aid of the Lagrange multiplier (LM) with four lags, which is asymptotically distributed, as x2 distribution. The use of dummy variables for the two depreciations of the drachma in the 1980-89 period does not alter the above results. * Indicates a significance level of 5 per cent. ** Indicates a significance level of 1 per cent.

to return to equilibrium; the error-correction term represents precisely this deviation. The term is expected to have a negative value, since output opposite will happen if it falls short of its equilibrium level. Finally, the size of the error-correction term reflects the speed at which the economy approaches long-run equilibrium. The error-correction term has the following general form:

significant long-term positive effect on production. Specifically, a real appreciation of 10 per cent brings about a 4.5 per cent rise in domestic

Î

¢y = Ó

™ i=1

Ï

·¢y(–i) +

™ i=1

Ì

‚¢m(–i) +

™ i=1

+ ™ ‰¢q(–i) + ™ Ë¢g(–i) + e0u(–1) i=1

Á¢w(–i) +

Ù i=1

(12)

output. This result suggests that the effects on the aggregate supply side prevail in the determination

where Â0 is the coefficient on the residuals from

of GDP over the long term.

the long-term relationship, u, while the other vari-

ECONOMIC BULLETIN, 14 12/99

13

ables are the first differences of the corresponding

production on imported raw materials, a fact

variables included in the long-run relationship.

which, in the long term, enhances the positive

These first differences represent the short-run

impact of an appreciation through the aggregate

effects. The value of the coefficient on the residu-

supply channel.

als, which ranges between –1 and 0, represents the speed at which real GDP adjusts to long-run equilibrium.

4. Conclusions

The results from the estimation of equation (12)

This study examined the effect of changes in the

are presented in Table 2.b. Among the explanatory

drachma’s real exchange rate on real domestic

variables, the real exchange rate is statistically sig-

product. The theoretical model developed distin-

nificant, with a time lag of four periods (one year)

guishes two transmission mechanisms, through

and with a negative sign. In addition, the change in

which the real exchange rate affects production.

real GDP, with a time lag of four periods, is statis-

The first works through aggregate demand and

tically significant, with a positive sign. The error-

has a negative effect, since an appreciation of the

correction term is statistically significant, and the

exchange rate leads to a drop in real GDP. The

size of Â0 indicates that more than half (52 per cent)

second works through aggregate supply and has a

of the deviation from the long-run equilibrium is

positive effect. Therefore, the final result depends

corrected in the current time period.

on the relative strength of the two alternative transmission mechanisms, which means that an

In general, the results of the estimation establish

appreciation of the real exchange rate is compati-

the view that the real appreciation of the drachma,

ble both with an increase and with a decrease in

while leading to a short-run decline in real GDP,

economic activity.

has, in the end, a positive effect, which offsets the initial decline. As a result, output increases as the

The results of the analysis show that the contrac-

economy moves to the long run.9 On the basis of

tionary effect of a real appreciation of the drachma

the preceding theoretical analysis, this means

on the economy’s output is temporary. The decline

that, in the short run, the effects from the aggregate demand prevail, while the effects from the supply side appear stronger in the longer run.10 We should finally note that the long-run exchange rate coefficient is, in absolute terms, almost double the short-term coefficient. This is largely due to two main features of the Greek economy: first, the rather low price elasticity of both imports and exports, which limits the negative short-term effect of the appreciation on aggregate demand.11 Second, the significant dependence of domestic

14

9 This result is in line with the conclusions of previous studies, according to which a devaluation only temporarily improves the external sector. See Brissimis and Leventakis (1992), Athanassoglou and Zombanakis (1992) and Karadeloglou (1990). 10 It should be noted that the results of the analysis do not change significantly if we use unit labour costs to estimate the real exchange rate. However, when we use the wholesale price index, while the long-term results do not differ, the real exchange rate appears statistically insignificant in the short term. This can be attributed, to a large extent, to the fact that this particular index does not include services and, therefore, does not provide any useful information on the interpretation of the relevant developments in an economy such as that of Greece, where a high percentage of GDP is generated by the services sector. 11 On the elasticities of Greece’s external trade, see Brissimis and Leventakis (1992), Garganas (1992) and Athanasoglou and Zombanakis (1992).

ECONOMIC BULLETIN, 14 12/99

The real exchange rate and economic activity

is mainly due to the decrease in aggregate demand

hard-drachma policy followed in recent years in

stemming from the deterioration in the current

Greece as a key tool in the effort to contain infla-

account balance; this, in turn, is a direct conse-

tion appears not to have had a contractionary effect

quence of the worsening of the international com-

on economic activity. The analysis suggests that

petitiveness of domestic products. Thereafter,

this policy can have a positive effect on output,

however, the favourable effect of the appreciation

notwithstanding the opposing views, which have

on production costs, through the reduced costs of

been expressed over the years. This result is not

imported raw materials, has an expansionary

surprising in the case of a small economy like that

effect on aggregate supply. This leads to a recov-

of Greece, in which the relatively low degree of

ery of output in the medium term.

substitution between domestic and foreign products is combined with a significant dependence of

In conclusion, the above analysis shows that the

ECONOMIC BULLETIN, 14 12/99

domestic production on imported raw materials.

15

References Alexander, S.S. (1952), “Effects of a devaluation on a trade balance”, IMF Staff Papers, 3, 263-78.

ment with reference to Greece”, Journal of Policy Modelling, 11, 247-71. Dornbusch, R. (1976), “Expectations and Exchange Rate Dynamics”, Journal of Political Economy, 84, 1161-76.

Alexander, S.S. (1959), ‘’Effects of a devaluation: ∞ simplified synthesis of elasticities and absorption approaches”, American Economic Review, 3, 23-42.

Edwards, S. (1986), “Are devaluations contractionary?”, Review of Economics and Statistics, August, 501-8.

Athanassoglou, P.P. and G.A. Zombanakis (1992),

Engle, R.F. and C.W. Granger (1987), “Cointegration

“External Debt and Macroeconomic Policy in a Small

and error correction: representation, estimation and

Open Economy: The Case of Greece”, in External

testing”, Econometrica, 55, 251-76.

Imbalances and Policy Constraints in the 1990s, Eds C. Milner and N. Snowden, The Macmillan Press Ltd.

Garganas, N. C. (1992), The Bank of Greece Econometric Model of the Greek Economy, Bank of Greece,

Argy, V. and J. Salop (1983), “Prices and output effects

Athens.

of monetary and fiscal expansion in a two-country world under flexible exchange rates”, Oxford Economic

Halpern, L. and C. Wyplosz (1997), “Equilibrium exchange

Papers, 35, 228-46.

rates in transition economies”, IMF Staff Papers, 44, 430-61.

Bank of Greece (1998), Monetary Policy, Interim Report 1998, November 1998.

Hanson, J. (1983), “Contractionary devaluation, substitution in production and consumption and the role

Bank of Greece (1999), Monetary Policy 1998-1999,

of the labour market”, Journal of International Econ-

March 1999.

omics, 14, 179-89.

Blanchard, O. J. and A. P. Muel (1993), “Competitive-

∫aradeloglou, P. (1990), “On the Existence of an

vess through disinflation: an assessment of the French

Inverse J-curve”, Greek Economic Review, 12, 285-305.

macroeconomic strategy”, Economic Policy, 16, 11-56. Khan, M. S. and M.D. Knight (1981), “Stabilisation proBranson, W.H. (1986), “Stabilisation, stagflation and

grammes in developing countries”, IMF Staff Papers, 28,

investment incentives: The case of Kenya 1975-80”, in

1-53.

Economic Adjustment and Exchange Rates in Developing Countries, ∂ds S. Edwards and L. Ahamed,

Krugman, P. and L. Taylor (1978), “The contractionary

University of Chicago Press, Chicago.

effects of devaluation”, Journal of International Economics, 8, 445-56.

Brissimis, S. N. and A. Leventakis (1989), “The effectiveness of devaluation: A general equilibrium assess-

16

Lizondo, J.S. and P. Montiel (1989), “Contractionary

ECONOMIC BULLETIN, 14 12/99

The real exchange rate and economic activity

devaluation in developing countries: An analytical

research project, University of Crete, Rethymmon, 11-13

overview”, IMF Staff Papers, 36, 182-227.

September 1998.

Machlup, F. (1966), International Monetary Economics,

Rhodd, R.G. (1993), “The effect of real exchange rate

George Allen & Unwin Ltd, London.

changes on output: Jamaica’s devaluation experience”, Journal of International Development, 5, 291-303.

Marston, C.R. (1985), “Stabilisation Policy in Open Economies”, in Handbook of International Economics,

Sachs, J.D. (1980), “Wages, Flexible Exchange Rates

vol. 2., Eds R. W. Jones and P.B. Kenen, North Holland,

and Macroeconomic Policy”, Quarterly Journal of

Amsterdam.

Economics, 94, 731-47.

Mitchell, A. and E.J. Pentecost (1998), “Exchange rate

Tsiang, S.C. (1959), “The role of money in trade bal-

policies, prices and supply response under transition”,

ance stability: Synthesis of the elasticity and absorption

Paper presented in a conference on the Phare-ACE

approaches”, American Economic Review, 51, 912-36.

ECONOMIC BULLETIN, 14 12/99

17

18

ECONOMIC BULLETIN, 14 12/99

Leading inflation indicators for Greece: an evaluation of their predictive power1

1. Introduction If the 1980s were the decade of global inflation upswings, the 1990s can be labelled as the decade of disinflation. Central banks in most OECD countries have adopted a particular inflation rate as their monetary target. In some cases, this target is accompanied by the attainment of exchange rate stability within the framework of official agreements, such as the Exchange Rate Mechanism of the European Monetary System. Since the mid-1990s, the main focus of macroeco-

Heather D. Gibson

nomic policy in Greece has shifted towards con-

Economic Research Department

taining inflationary pressure through an exchange rate target. The objective of this policy is to reach

Sophia M. Lazaretou

an inflation level compatible with that of current

Economic Research Department

inflation levels in the euro area. Specifically, the Bank of Greece aims at reducing inflation to a level consistent with the Maastricht criterion for the drachma’s entry into the euro area. Most OECD member countries have been following successful anti-inflationary policies since the mid1980s. In Greece, by contrast, inflation remained very high for a period of over 20 years, often exceed-

1 We wish to thank Joanna Bardakas, Panayotis Kapopoulos, Isaac Sabethai and Nicos Zonzilos for their helpful comments and discussions. We especially thank George Tavlas for his constructive comments, which proved essential in improving the paper, and for his support. We also thank the participants in the Project LINK Fall Meeting, Athens, 1-5 November 1999, for their comments. Finally, we thank the Money and Banking Division and the Domestic Economy Division of the Economic Research Department of the Bank of Greece for providing their data. Responsibility for any mistakes or omissions lies wholly with the authors, while the views presented in this paper do not necessarily reflect those of the Bank of Greece. This paper is based, to a great extent, on a previous research paper entitled “Leading Inflation Indicators for Greece”, Economic Modelling, forthcoming.

ECONOMIC BULLETIN, 14 12/99

19

ing 24 per cent.2 Only in 1995 did inflation fall to

high. In the early 1990s, however, this trend was

single digits, for the first time since 1972. From that

reversed and Greek inflation has been gradually

year onwards, a period of rapid disinflation began;

converging towards EU levels.

the goal is to reduce inflation below 2 per cent. Chart 3 depicts the relationship between inflation The persistence of strong inflationary pressures

and the GDP growth rate per employee in differ-

over the past twenty years is in stark contrast with

ent periods. We observe that in periods of infla-

the monetary stability of the 1960s, when infla-

tionary peaks (1970s and 1980s) the relationship

tion was below 5 per cent. Indeed, as shown in

is strongly negative.

Chart 1, the average annual change in the Consumer Price Index in the decade 1960-1969 was

The effort made by the financial authorities to fol-

1.95 per cent, while in the period 1970-88 the

low an anti-inflationary policy rekindled resear-

rate of change was eight times higher (15.9 per

chers’ interest in developing methods to predict

cent). In the period 1990-98, inflation was at sig-

the inflationary process. These methods can be

nificantly lower levels.

divided into four categories. First, predictions about future inflation levels are made through the estima-

Chart 2 shows the difference in terms of inflation between Greece and the 15 member states (Greece inclusive) of the European Union. We note that, from the mid-1970s onwards, the deviation of Greek inflation from the average EU rate was very

20

2 For a detailed review of the factors explaining the inflation phenomenon in postwar Greece and its main characteristics, see Leventakis and Brissimis (1980), Sarantis (1982), Alogoskoufis (1986, 1995), Kalyvitis and Lazaretou (1997), Demopoulos and Kapopoulos (2000).

ECONOMIC BULLETIN, 14 12/99

Leading inflation indicators for Greece

tion of a structural economic model. Second, the

research, or indirectly, by studying the behaviour

evaluation of inflationary expectations of market

of asset prices, provides useful information about

agents, either directly, via surveys and market

the inflationary process. By incorporating this infor-

ECONOMIC BULLETIN, 14 12/99

21

mation into their decision-making, economic agents

Our results suggest that only six variables have

can avoid systematic errors. Third, the drawing up

the characteristics of a leading indicator of infla-

of various macroeconomic indices and the study of

tion. Four of them (import prices, wholesale prices,

their historical development may provide some

money growth and world commodity prices) have

indications about the future course of inflation and

a short average lead time of about 3 months, while

allow the formulation of ‘rules of thumb’. Fourth,

the other two (retail sales volume and industrial

there has recently been a revival of interest in the

production) have a longer average leading time of

application of the leading inflation indicator

about 14 months. On the basis of leading times,

approach, which draws upon the methodology

we construct two composite indices, a longer and

established by Burns and Mitchell (1946) and

a shorter one, that combine the information incor-

recently revised by Lahiri and Moore (1991) for

porated in the individual variables, and we evalu-

the identification and study of business cycles. We

ate their ability to predict inflation upswings and

accept the hypothesis that inflation, like output, is

downswings. Empirical findings reveal that both

cyclical, with alternating peaks and troughs. We

composite indices perform quite well as leading

examine the behaviour of a wider set of macroeco-

indicators of inflation turning points, i.e. the num-

nomic variables in order to determine if their turn-

ber of inflation turning points they miss or falsely

ing points precede those of inflation, so that we can

signal is very limited.

use these variables as leading indicators. It should be noted that the leading indicator approach does

The remainder of the paper is organised as fol-

not aim at forecasting the level of future inflation.

lows. Section 2 presents the typical characteristics

Rather, it aims to identify in time future turning

of the cyclical behaviour of the inflationary

points of inflation and is, therefore, considered sup-

process in Greece. Section 3 describes individual

plementary to other forecasting methods.

and composite leading indicators. Section 4 evaluates the performance of these indicators, i.e.

This paper looks for the existence of leading inflation

their correlation with inflation swings and their

indicators for Greece, that is, of economic variables

ability to identify, or even predict, inflation turning

or composite economic indices that incorporate the

points. Finally, Section 5 summarises the main

joint effect of individual variables. The definition

conclusions.

of these variables is valuable for a timely warning about the appearance of significant changes in the behaviour of the inflationary process. Specifically, we examine the predictive power of 20 macro-

2. The cyclical behaviour of inflation in Greece

economic series, using monthly data for the period 1954-1998. From the overall set of economic vari-

We use the twelve-month rate of change in the

ables, we attempt to uncover those variables whose

Consumer Price Index (CPI)3 as our measure of

changes in behaviour usually precede changes in the behaviour of inflation. Therefore, these variables can contribute to a timely warning about the appearance of future turning points in the inflation cycle.

22

3 The core inflation time series is considered smoother and, therefore, more conducive to finding “real” turning points. Unfortunately, the available data do not cover a long period; that is why this time series is not used in the paper.

ECONOMIC BULLETIN, 14 12/99

Leading inflation indicators for Greece

inflation over the period from January 1954 to

difference between a trough and a peak, cannot

December 1998. Chart 4 depicts the inflation time

be smaller than 1.5 percentage points.

series, along with inflation cycles. The shaded areas represent the downward phases of inflation. The

(d) The length of time over which the change from

peaks and troughs of the cycle are represented by

an upswing to a downswing, or vice versa, takes

the left-hand and right-hand edge, respectively, of

place, must be at least six months.5

each shaded area. (e) Finally, between two or more turning points The identification of peaks and troughs confirms

with equal values, we select the most recent

that inflation, like output,4 is cyclical. To deter-

point.

mine inflation cycles we use five criteria, commonly found in the literature (see, for example,

Table 1 provides detailed information on inflation

McNees, 1991; Roth, 1991; Artis et al., 1995):

cycles identified on the basis of the above-men-

(a) Peaks are always followed by troughs and, conversely, troughs are always followed by peaks. (b) The turning point is the most extreme point between upswings and downswings. (c) The size of the change, namely the absolute

ECONOMIC BULLETIN, 14 12/99

4 For a study of output cycles in the Greek economy, see Dongas (1992). 5 To define the upswing and downswing phases, we test various minimum lengths for the cycles: 6, 9 and 12 months. By choosing a 9-month cycle, we miss two cycles (1956-57 and 1975-76). Choosing a 12-month cycle we miss an additional cycle (199293). Since the amplitude of these cycles is greater than the 1.5 percentage-point minimum and therefore meets the fourth criterion, we cannot exclude them. We therefore choose a 6-month period as the minimum duration of an inflation cycle.

23

Table 1 Turning points of Greek inflation (1955-1998) Peak

Date

Trough Percentage change

Date

Percentage change

Amplitude of percentage change

Duration (months)

Contraction

Contraction

Expansion

Expansion

P1: 1956/2

7.342

T1: 1957/2

0.684

-6.658 (-0.555)

-

12

-

P2: 1957/10

3.546

T2: 1958/11

0.564

-2.982 (-0.229)

2.862 (0.358)

13

8

P3: 1961/3

4.118

T3: 1962/2

-2.083

-6.201 (-0.564)

3.554 (0.127)

11

28

P4: 1963/3

4.415

T4: 1964/4

0.367

-4.048 (-0.311)

6.498 (0.500)

13

13

P5: 1966/4

5.914

T5: 1968/4

-1.471

-7.385 (-0.308)

5.547 (0.231)

24

24

P6: 1974/3

33.470

T6: 1975/7

10.704

-22.766 (-1.423)

34.971 (0.492)

16

71

P7: 1976/6

14.818

T7: 1977/3

10.802

-4.016 (-0.446)

4.114 (0.374)

9

11

P8: 1980/6

26.997

T8: 1985/7

16.634

-10.363 (-0.135)

16.195 (0.415)

61

39

P9: 1986/7

24.710

T9: 1988/6

11.749

-12.961 (-0.564)

8.076 (0.673)

23

12

P10: 1990/11

23.889

T10: 1992/7

13.549

-10.340 (-0.517)

12.140 (0.419)

20

29

P11: 1993/5

16.431

2.882 (0.288)

-

10

9.684 (0.388)

20.2

24.5

Average

-8.772 (-0.505)

Note: Inflation is measured as a twelve-month percentage change in the non-seasonally adjusted Consumer Price Index (1994=100). The identification of each turning point was done on the basis of the five criteria mentioned in the text. The average monthly change is calculated as the mean absolute change over months classified as periods of inflation contraction or inflation expansion and appears in parentheses in the column headed “amplitude of percentage change”.

tioned criteria. Four features of the data are espe-

duration of inflation is evident.6 Upswings lasted

cially worth noting. First, over the period under

an average of two years (24.5 months), during

study, we identify 10 inflation cycles. Second,

which the average rise in inflation was 0.4 per-

there are significant differences in amplitude

centage point per month. Downswings were

between cycles. During upswings, inflation rose

slightly shorter. Their average duration did not

by an average of 9.7 percentage points, while dur-

exceed 20 months and the average fall in inflation

ing downswings it fell by an average of 8.8 per-

was 0.5 percentage point per month. The steepest

centage points. The variance between the differ-

fall (–1.4 percentage points per month) occurred

ent turning points was also quite significant. In

between March 1974 and July 1975, when the

downswings, the absolute change varies from a

government, in an attempt to control inflationary

high of 23 percentage points to a low of 3 per-

pressures (the annual inflation rate had peaked at

centage points. Similarly, in upswings, it varies

over 33 per cent in the aftermath of the collapse of

from a high of 35 percentage points to a low of 3

the Bretton Woods system and the first oil crisis),

percentage points.

imposed strict price and wage controls. The steepest rise (0.7 percentage point per month) occurred

Third, the duration of cycles varies significantly. Specifically, inflation downswings vary from 9 to 61 months and inflation upswings from 8 to 71 months. In any case, a slight asymmetry as to the

24

6 Elements of this asymmetry are evident in inflation cycles for other European economies. See, for example, Artis et al. (1995) on the UK, Bikker (1993) on Holland, and Cabrero and Delrieu (1996) on Spain.

ECONOMIC BULLETIN, 14 12/99

Leading inflation indicators for Greece

between July 1985 and July 1986, resulting from

Greece’s discount rate and the 12-month Treasury

the devaluation of the drachma in October 1985.

bill rate, and we evaluate their predictive power.

Fourth, turning points seem to follow a seasonal

The second category includes cost and price vari-

pattern. Most of them (6/10 troughs and 8/11

ables. Assuming a constant mark-up, increases in

peaks) occur in the first half of each year. This

input prices (raw materials, labour costs) are shifted

behaviour may reflect the fact that most changes

onto output prices, thus causing a rise in inflation

in taxes and public utility rates usually take place

(Clark 1995). We consider as leading indicators

in the first half of each year.

the time series of world commodity prices,9 oil prices, the exchange rate, wholesale prices (the Wholesale Price Index and its main components),

3. Leading inflation indicators

raw material prices and unit labour costs.10 We should note, however, that the usefulness of these

Having presented a turning-point chronology for

variables as leading indicators of inflation can be

inflation in Greece, we attempt in this section to

reduced quite drastically in the case that wage

identify leading inflation indicators, i.e. economic

rises are in step with increased productivity or

variables, which, either on their own or in the

when the mark-up is not constant11 (Clark 1995).

form of a composite index, allow us to predict turning points in inflation. Although our analytical

The third category refers to those variables that

approach is largely statistical, economic theory does provide us with a host of candidate indicators that may influence, directly or indirectly, the behaviour of the inflationary process. Specifically, we gather information from three broad categories of economic aggregates.7 The first includes the variables reflecting monetary developments in the economy, such as the money supply and interest rates. If we accept the hypothesis that inflation is a monetary phenomenon, then the main monetary variables, such as the money supply (both narrow and broad money), credit expansion and interest rates should be considered potential leading indicators. However, the use of some of these variables as policy instruments significantly reduces their effectiveness as indicators of inflation.8 As Table A3 in the appendix shows, we consider as candidate leading indicators the rate of growth of M0, M3 and M4, total credit expansion, along with the Bank of

ECONOMIC BULLETIN, 14 12/99

7 For an extensive presentation of economic aggregates as leading inflation indicators, see Gibson and Lazaretou (1999). Table A3 of the Appendix presents the variables used in the analysis (definition, source, sample size). 8 Example 1. We assume that the central bank uses a short-term interest rate as a tool of anti-inflationary policy. If it anticipates a rise in inflation, it will tighten monetary policy now by raising interest rates, hoping to contain inflationary expectations. If the policy is successful, then inflation will not rise in the immediate future. In other words, interest rates, as monetary policy tools, can smooth out inflation cycles, with the result that there appears not to be any relationship between inflation and interest rates. Example 2. Suppose that current inflation (t) is determined by the previous period’s inflation (t-1) and monetary expansion (mt-1), i.e. t = Út-1 + mt-1 + Ât, where Ât is a random variable. If the central bank changes the rate of growth of the money supply, so as to achieve zero inflation in the next period, that is, if Ett = 0 or Út–1 = -mt-1, then current inflation (t) is not related to the rate of monetary growth, i.e. t = Ât. 9 See Webb and Rowe (1995) and Garner (1995) on the usefulness of world commodity prices in predicting US inflation. 10 There are no monthly figures available on wages or unit labour costs in the Greek economy. As a result, these variables are not included in the analysis. 11 It has been empirically demonstrated that in small, open European economies, the mark-up is not constant (see, for example, Dombrecht and Moes 1998, and, for Greece, Zonzilos 1999). Thus, a small increase in input prices does not necessarily lead to a one-for-one increase in output prices, and hence in CPI inflation.

25

reflect excess demand or supply conditions. Under

ber of months every time and (b) this particular

conditions of excess demand, when aggregate

variable does not send out any false signals, that

demand continues to rise unabated and approaches

is, it neither forecasts a turning point in inflation

capacity output, inflationary pressures are expected

which does not then occur, nor does it fail to fore-

to reemerge in the immediate future. Therefore,

cast any actual turning points. Therefore, identify-

turning points in variables such as capacity utilisa-

ing the economic variable’s turning points in rela-

tion, and the gap between current and potential out-

tion to those of inflation is the first step in assess-

put may be leading indicators of inflation (Artis et al.

ing the usefulness of a variable as a leading infla-

1995, Webb and Rowe 1995). In the present analy-

tion indicator.14 For this purpose, we obtain a total

sis, we use industrial production and retail sales vol-

of 20 macroeconomic variables related to the

ume as measures of demand pressure. Variables

Greek economy (see Table A3 in the Appendix)

such as employment, the unemployment rate and

and we identify the turning points of their cycles,

vacancies also provide useful information on supply

always in relation to the inflation cycle. All time

and demand conditions in the labour market and, by

series except interest rates are expressed as 12-

extension, on inflationary developments. This is

month growth rates. To eliminate “erratic” short-

because a high rate of employment growth signals a

term fluctuations that may appear in some series,

future rise in inflation, since high demand for labour

we smooth those affected by taking a small num-

intensifies competition between enterprises for per-

ber of periods (less than a year).15

sonnel and raises wages.12 Of the 20 macroeconomic variables examined, Finally, variables reflecting inflationary expectations

only six provide some leading indicator informa-

may provide useful information about the future

tion. Even though the turning points of these vari-

course of inflation. For example, an estimation of

ables are mostly determined with the help of ad

inflationary expectations can be made on the basis

hoc techniques and through a visual inspection,

of either information derived from market surveys

the need to adopt an objective criterion is obvi-

(Roth 1991, Bikker 1993, Garner 1995), or of

ous. Thus, the variable we finally choose as a lead-

developments in asset prices (Fagan 1996). Such a

ing indicator should satisfy the following criteria:

measure is the slope of the yield curve, which is a

(a) its cycle should, on average, lead that of infla-

graphic representation of the relationship between

tion; (b) there should be little variance around

securities’ yield and their maturity (Webb and

leading points; (c) the number of inflation turning

Rowe 1995, Cabrero and Delrieu 1996, Davis and Fagan 1997 and Baumgartner et al. 1997).13

4. Inflation cycles and leading indicators An economic variable is a leading indicator of inflation if (a) the turning points of its cycle precede those of the inflation cycle by the same num-

26

12 In the case that wage rises outpace productivity gains. 13 We do not test whether asset prices are a leading inflation indicator. This is because, as Gibson and Brissimis (1999) found, the yield curve provides very little information about expected inflation in Greece. 14 To identify the cycles for each economic variable, we used precisely the same criteria as employed in identifying inflation cycles. 15 The erratic movements are obvious by simply observing the time series charts. Detailed information about how the series were smoothed is provided in Table A3 in the statistical appendix. See also Webb and Rowe (1991) and Artis et al. (1995).

ECONOMIC BULLETIN, 14 12/99

Leading inflation indicators for Greece

Table 2 Correlation with the turning points of the inflation cycle [number of lead (-) or lag (+) months] Inflation turning points: Peak (P), trough (T) P1: 1956/2 T1: 1957/2 P2: 1957/10 T2: 1958/11 P3: 1961/3 T3: 1962/2 P4: 1963/3 T4: 1964/4 P5: 1966/4 T5: 1968/4 P6: 1974/3 T6: 1975/7 P7: 1976/6 T7: 1977/3 P8: 1980/6 T8: 1985/7 P9: 1986/7 T9: 1988/6 P10: 1990/11 T10: 1992/7 P11: 1993/5

RS (1)

IP (2)

M0 (3)

WCP (4)

IMP (5)

WSPID (6)

na na na na na na na na m -9 -36 -8 +10 m m -34 -21 -18 -25 -12 m

na na na na -2 +4 m m -20 -3 -4 -5 +7 +11 -12 -30 -6 -6 -24 -15 -13

+8 +7 +11 +12 -1 0 -3 -5 -17 -12 -1 -11 -11 -11 -20 -17 -6 -3 -7 -1 -1

na na na m -14 -14 m m +3 -9 -7 -4 +9 +12 +11 -2 m m -29 -31 +16

na na na na na na na na na na -1 -1 0 +2 +9 m -7 -19 -6 m m

na na na na na na na na na 0 0 +7 m m 0 -31 -7 -16 -1 m m

Number of extra turning points

0

3

7

4

2

0

Number of missing turning points

4

2

0

5

3

4

8/13

12/17

17/21

8/18

6/11

7/12

9

18

28

17

10

8

-17.0 -18.0 -16.2

-7.9 -9.2 -6.3

-4.2 -4.4 -4.1

-4.5 -1.6 -8.0

-2.9 -1.0 -6.0

-6.0 -2.0 -10.0

Standard deviation of leads and lags (No. of months) All points Peaks Troughs

13.5 17.1 9.6

10.9 9.4 12.3

8.8 8.9 8.6

14.4 14.8 13.0

7.6 5.7 9.3

11.3 2.9 14.7

Coefficient of accuracy (k)

0.04

0.04

0.06

0.02

0.04

0.05

Frequency of leads/number of inflation turning points Number of turning points of each indicator Average lead or lag (No. of months) All points Peaks Troughs

Note: Extra turning points: (i) M0 money growth, 1969/7 (T), 1977/4 (P), 1978/3 (T), 1981/4 (T), 1982/10 (P), 1995/1 (T), 1997/9 (P), (ii) industrial production, 1966/2 (T), 1966/10 (P), 1993/9 (T), (iii) import prices, 1982/5 (T), 1997/2 (T) and (iv) world commodity prices,1969/8 (P), 1971/11 (T), 1981/9 (T), 1983/10 (P). The coefficient of accuracy is defined as k=(1/Ûi)x(Li2/(TRFxTi)) i=1,...,20 indicators, where Ûi is the standard deviation of i indicator, Li and Ti are, respectively, the number of leads and the number of total turning points of each indicator, and TRF is the number of inflation turning points. High values of k indicate higher accuracy. m: missing turning point, na: non-available data.

points that were either not identified in time or

these criteria, we reject those variables that have

were wrongly identified as such should be kept to

no clear turning points, or have turning points that

a minimum; and (d) there should be a high degree

do not generally precede inflation turning points,

of accuracy in the sense of a large number of iden-

or either miss too many inflation turning points or

tified turning points and less dispersion. Based on

generate too many false signals.

ECONOMIC BULLETIN, 14 12/99

27

Table 2 provides detailed information about the

describe the real economy. The other four, import

cycles of the 6 finally accepted indicators, in rela-

prices (IMP), the wholesale price index (WSPID),

tion to the inflation cycle. Two variables, indus-

the money supply (M0), and world commodity

trial production (IP) and retail sales volume (RS)

prices (WMD), are nominal indicators. Charts 5-10

28

ECONOMIC BULLETIN, 14 12/99

Leading inflation indicators for Greece

illustrate movements in the 6 variables selected in

inflation, as is evident from the predominance of

relation to inflation cycles.

minus signs. For example, the growth rate of the money supply (M0) turns downwards one month

We observe that the indicators, on average, lead

ECONOMIC BULLETIN, 14 12/99

before the inflation peak in March 1961. On the

29

other hand, industrial production has a trough

ever, money growth has the highest incidence of

30 months earlier than the inflation trough in

false signals, in contrast to retail sales volume

July 1985. All variables, except money growth,

and wholesale prices, which have no false sig-

miss two or more inflation turning points. How-

nals.

30

ECONOMIC BULLETIN, 14 12/99

Leading inflation indicators for Greece

The lower half of Table 2 shows the average num-

an average of 2.9 months and retail sales by 17

ber of months that an indicator leads (–) or lags

months. Moreover, the high standard deviation

(+) an inflation turning point, as well as the stan-

reveals a degree of instability in the relationship

dard deviation. For example, import prices lead by

between the turning points of each series and

ECONOMIC BULLETIN, 14 12/99

31

point, then, although we may be quite sure that

Table 3 Inflation composite leading indicators: Shorter Indicator (SLI), Longer Indicator (LLI) Inflation turning points: Peak (P), Trough (T) P1: 1956/2 T1: 1957/2 P2: 1957/10 T2: 1958/11 P3: 1961/3 T3: 1962/2 P4: 1963/3 T4: 1964/4 P5: 1966/4 T5: 1968/4 P6: 1974/3 T6: 1975/7 P7: 1976/6 T7: 1977/3 P8: 1980/6 T8: 1985/7 P9: 1986/7 T9: 1988/6 P10: 1990/11 T10: 1992/7 P11: 1993/5

inflation will turn at some point in the future, we cannot be sure exactly when. Based on the average leading time, we can classify the 6 individual indicators into two categories. The first includes four of the 6 variables (money growth, wholesale

SLI

LLI

prices, world commodity prices, import prices),

+7 m m +12 m m -3 -8 -17 +10 -1 +6 m m -4 -7 -7 -17 -2 -3 -5

na na na na -2 +4 m m -20 -5 -9 -8 m m -20 -33 -19 -16 -26 -14 m

Number of extra turning points

0

2

Number of missing points

6

5

Frequency of leads/number of inflation turning points Average lead or lag (No. of months) All points Peaks Troughs

11/21 (52%)

11/17 (65%)

with a shorter average lead time (3-6 months). The second contains the other two variables (industrial production and retail sales volume) with a longer average lead time, of around 8 and 17 months, respectively. The combined effect of the individual indicator in each category is reflected in the composite leading indicators of inflation, a shorter (SLI) and a longer (LLI) one. To construct these indices, we first adjust each series for differences in levels and volatility, by subtracting its means and dividing by its standard deviation. Second, on the assumption that each component has the same weight, the composite index is the simple average of the individual variables in each category. Table 3 and Charts 11 and 12 provide information about the future behaviour of inflation on the

-2.6 -4.0 -1.0

-14.0 -16.0 -12.0

basis of the composite indices. We observe that, first, the index cycles lead inflation cycles, on average; this is confirmed by the high incidence of the

Standard deviation of leads and lags (No. of months) All points Peaks Troughs

minus signs. Average lead time of the shorter 8.3 6.3 9.9

10.1 8.0 11.4

Note: The LLI indicator has two extra points, 1977/3 (P) and 1978/2 (T). m: missing turning point, na: non-available data.

those of inflation. The dispersion in lead time frequency16 significantly limits the effectiveness of the variables as leading indicators of inflation. In other words, if the indicator signals a turning

32

index is about 3 months, while that of the longer index is over a year. Second, the mean lead for inflation peaks is greater (4 months for the SLI and

16 The great dispersion of leads and lags, and the number of inflation turning points that were either missed or falsely signalled, are consistent with the results of similar studies concerning other countries (see, for example, Artis et al. 1995 for Great Britain, Roch 1991 for the US, and Bikker 1993 for the Netherlands).

ECONOMIC BULLETIN, 14 12/99

Leading inflation indicators for Greece

16 months for the LLI) than for inflation troughs (1

including period t and if the previous regime is

month and 12 months, respectively). Third, as in

an upswing, then the indicator signals at time t

the case of the individual indicators making up the

a peak in inflation. In more formal terms, this rule

indices, the dispersion in the frequency of the

is expressed as follows:

leads, as measured by standard deviation, is quite large, even though the SLI is less volatile than the

(i) If ¢It–1>0, ∀i=0,1,2, and if the previous regime

LLI.

is a downswing, then I signals at time t that a trough occurred.

Finally, the frequency of false signals is significantly reduced, compared with the individual compo-

(ii) If ¢πt–1