inventories and the business cycle

THE ECONOMIC RECORD, V O L 71. NO. 212. MARCH 199s. 27-39

Inventories and the Business Cycle* DARREN FLOOD and PHILIP LOWE

Economic Group, Reserve Bank of Australia,

Sydney, NSW 2000

This paper examines the relationship between the inventory cycle and the business cycle using both macroeconomic and survey data It is argued that over the past decade and a half; the changes in inventory management have reduced the amplitude of the inventory cycle. The paper also argues that the behaviour of inventories is consistent with demand shocks being an important source of business cycle /ructuations.

I Introduction On average, over the past 30 years, inventory

investment has accounted for just 0.7 per cent of the level of Australian Gross Domestic Product (GDP). In contrast, on average, quarterly changes in the level of inventory investment have accounted for 59 per cent of the change in GDP. In recessions, inventory investment has played an even more important role, often accounting for over 100 per cent of the fall in output. Clearly, while the accumulation of inventories has little impact on long-run economic growth, it is an extremely important part of the business cycle. This importance has led to a considerable research effort by both policy makers and businesscycle theorists into the behaviour of inventories. This research has failed to arrive at a consensus. Par- allelling the ongoing debate concerning the source of shocks generating the business cycle, there are basically two broad schools of thought. One school argues that the inventory cycle is driven by

We arc grateful to Tom Rohling and Richard De Abreu Lourenco for helpful rescarch assistance and to two anonymous referees. a co-cditor of this journal and participants at the Reserve Bank Research Seminar for helpful comments. Any e m an OUIS alone. The views expressed herein arc those of the authon and do not necessarily reflect those of the Resave Bank of Australia.

shocks to demand while the other argues that i t is driven by shocks to supply.'

In this paper we examine the role of inventory investment in the Australian business cycle and use a previously unexplored dataset to examine the sources of fluctuations in inventories. The data are from the ACCWestpac Survey of Industrial Trends and consist of firms' expected and actual experience with inventories, costs and demand. Under certain conditions the survey results allow us to distinguish between expected and unexpected changes in these variables. Our results suggest that demand factors have an important role to play in driving the inventory cycle. In addition, they suggest that an increased ability of firms to manage their inventories has meant that unexpected changes in demand no longer generate the same pronounced inventory cycle that they once did. This conclusion is supported by the macroeconomic data

The demand side models of inventory investment were pioneered by the work of Metzler (1941). who argued that desired inventories were proportional to expected sales. This link between expected sales and inventories reflects two factors. The first is that marginal cost increases with output and the second is that tirms hold inventones to reduce the probability of being unable to

1 See Kahn (1992) for evidence that demand shocks dominate and West (1989) for evidence that cost (or supply) shocks dominate.

2a ECONOMIC RECORD MARCH

meet unexpectedly high demand. With increasing marginal costs, firms can minimize production costs by keeping a constant level of production. equal to expected sales. With production decisions being made before demand is known, a positive unanticipated shock to demand is met, in the first instance, by running down inventories. This fall in inventories, however, increases the probability of running out of stock and so leads to some additional increase in output in the subsequent period to rebuild stocks.

In contrast. the supply-side models focus on cost shocks. These models are based on the premise that the principal type of shock generating the business cycle is a shock to the production function. When a favourable but temporiuy productivity shock occurs, it reduces costs and induces the firm to increase output while costs are temporarily low. This extra output is stored as inventories and sold when output is temporarily low as the result of a bad productivity shock.

In recent years, these supply-side models have received increased attention. This interest primar- ily reflects two stylized facts obtained from studies of the behaviour of inventories in the United States. The first is that the variance of production exceeds the variance of sales. The second is that inventory movements are pro-cycIical.2 Roponents of cost-based theories argue that the greater variance of production than sales is inconsistent with the demand shock-production smoothing model. The positive correlation between inventory investment and output is also seen to be inconsistent with this model.

Given the role of the inventory cycle in deter- mining the amplitude of the output cycle it is important to understand the factors driving inventory investment. This is the task of the remainder of this paper. We begin in Section U by briefly discussing the relationships among output. demand and inventories predicted by the various models. In the following section. we use macroeconomic data to present some basic facts about inventory investment in Australia. In Section IV.

* For US evidence on the relationship between rhe variance of production and the variance of sales see Blanchard (1983). Blinder (1986). West (1988) and Blinder and Maccini (1991). In contrast to the results for h e US. Bevon (1993) finds that for Japanese f m , the variance of production is typically less tban the variance of sales. For US evidence on the procyclical nature of inventory investment see Summen (1981), Blanchard and Fischer (1989) and West (1989).

we turn to the survey data to analyze the relationship between changes in demand and costs on the one hand, and expected and unexpected changes in inventories of finished goods and raw materials on the other. Finally, in Section V we summarize and conclude.

I1 Relationships between Demand Output and Inventory Investment

The two principal models of inventory investment-the demand shock-production smoothing model and the productivity shock model-predict quite different relationships between demand, output and inventories. Before examining the Australian data on these relationships, we briefly discuss the factors that affect the sign and strength of these relationship^.^

We begin by considering the implications of a temporary shock to the cost of production-yy, an improvement in productivity. With productivity temporarily high. the firm will wish to increase production. since costs are temporarily low. It will store this extra production and sell it at some future point when productivity is temporarily low. If we move from considering a positive productivity shock to a single firm. to an economy-wide shock. then the level of income and thus demand should also increase. Consequently, changes in output. demand and inventory investment should all be posirively correlated. Further, given that demand is a function of permanent rather than current income, the variability of production should be greater than the variability of demand.

The extent to which inventories increase in response to a productivity shock depends upon the firm’s need to smooth production. the cost of holding inventories, and the persistence of the productivity shock. The more rapidly marghal costs increase with output, for a given level of productivity. the smaller will be the increase in output and inventories. Similarly, the greater the cost of holding inventories (for example, interest costs, storage costs and depreciation) the smaller will be the output and inventory responses. Finally. the size of the inventory response is de

creasing in the persistence of the shock. The shorter the time period over which low production costs are likely to apply, the greater will be the inventory response. If the shock is permanent.

3 For a formal presentation of the ideas discussed in this section see Blanchard and Fischer (1989) pp.305-8.

I995 INVENTORIES AND THE BUSINESS CYCLE 29

there is no need to move production from future periods to the current period.

Now consider the case of an unanticipated increase in demand where production decisions must be made before demand is known. In the first instance, inventory investment will be negative, as the additional demand is met entirely from stocks. As a result, inventory investment and demand will be negatively correlated. In subsequent periods. the impact on output and inventories depends importantly on the rate at which marginal cost increases with output and the persistence of the demand shock. If marginal costs increase extremely rapidly with output, then the firm will wish to meet the additional demand arising from a persistent demand shock by contin- uing to run down inventories, rather than by increasing production. In the most extreme case, a temporary demand shock does not generate a business cycle, as it is completely offset by inventory investment.

On the other hand, if marginal costs do not increase with output, the output cycle is likely to be more pronounced than the demand cycle. Fol- lowing the demand shock. firms will wish to increase output., not only to meet the higher expected demand in subsequent periods, but also to reestablish some desired stocks to sales ratio. The higher is this desired ratio, the larger is the likely output response. Thus, improvements in inventory management techniques that reduce the target stocks to sales ratio should see unexpected changes in demand produce smaller inventory and output cycles.

Finally. we consider the relationships between inventories, output and demand when the increase in demand is expected, as opposed to unexpected. If production smoothing is important and the increase in demand is expected to last for a relatively short period of time, the firm will be willing to meet some of the extra demand out of inventories. As a result. output will rise by less than the increase in demand and inventories will fall. However, the more persistent is the increase in demand, the less willing the f m will be to meet the additional demand out of inventories. Simi- larly. if marginal costs are constant. increasing output to match the higher demand does not increase per unit output costs and the fkm will respond to the higher expected demand by increasing inventories to help maintain a constant stocks to sales ratio. In this case demand and inventories will be positively comlated.

In summary, if production decisions are made

before demand is known, unexpected changes in demand and inventory investment should be negatively correlated. In contrast, if changes to demand are expected, then demand and inventory investment can be either positively or negatively correlated, depending upon the rate at which marginal costs increase with output. Whether or not changes in demand are expected or unexpected, demand shocks are consistent with output being more variable than demand, provided that marginal costs do not increase too quickly with output. Finally, for both unexpected and expected changes in demand, the output response is increasing in the target ratio of stocks to sales.

III Inventory Investment in Australia Four major downturns in economic activity

have occurred since the beginning of the 1960s. Figure I illustrates the pattern of private non-farm inventory investment during these downturns. In each case the reference point 0 on the horizontal axis represents the trough in Gross Domestic Product (GDP).

It is clear from this graph that during the 1960s and early 1970s. inventory behaviour followed a quite marked cyclical pattern. In the two most recent recessions the pattern has been considerably less pronounced with inventory investment actually declining as economic growth slowed. Along with this change in the amplitude of the inventory cycle there has been a significant decline in the stocks to sales ratio (see Figure 2).

In their comprehensive survey of the US inventory investment literature, Blinder and Maccini ( I99 1) point to two stylized facts that any theory of inventory investment should be able to explain. The first of these stylized facts is that the variance of output exceeds the variance of sales4 As Table 1 shows, the Australian data are consistent with the US data; the ratio of the variance of output to the variance of sales is greater than one in all industries (column 3)? This evidence weighs against the basic production smoothing model. If f m s really do attempt to maintain production

Beason (1993) finds the opposite result for Japanese iodustries.

J These variances arc calculated from &trended series for sales and inventories. This is tbe proccdm used by Blinder and Maccini (1991). However, if the level of inventories is an integrated process. the decnnd- ing procedure may be invalid. Accordingly, we also calculate the variances of the log difference in sales and output. The results arc similiar.

30

%

3

2

1

0

-1

-2

ECONOMIC RECORD MARCH

FIGURE I Increase in Privcrre Non-Farm Srockr. Ck of Domesric Final Demund

Yo

3

2

1

0

-1

-2

Notes: 1. Data are seasonally adjusted and are in 1989MI prices. Domestic final demand is obtained by adding final consumption expenditure to gross fixed capital expenditure. All data are from Australian Bureau of Statistics (ABS) 5206.0.

2. The troughs in GDP are in September 1961, September 1975, June 1983 and September 1991 respectively. The choice of September 1975 is somewhat arbitrary since this period was characterized by slow growth rather than a clear recession.

constant in the face of stochastic sales, then production should be less variable than sales.

The fifth column of Table 1 provides Australian evidence on the second stylized fact identified by Blinder and Maccini; that is, changes in sales and stocks are positively, rather than negatively, correlated. The results show that the majority of the correlation coefficients between log changes in inventories and sales are positive, although only two of the coefficients are statistically different from zero. Similar results are obtained when the correlation coefficients between changes in detrended sales and detrended inventories are calculated.

This piece of evidence is often seen as dam- aging to demand shock driven models of inventones and the business cycle. As the discussion in Section II highlighted, if stocks are being used as a buffer against stochastic sales, the change in stocks should be negatively correlated with sales. However, in Section II it was also noted that if the changes in demand were expected then sales, output and inventories might reasonably be expected to be positively correlated. To answer the question of whether unexpected changes in demand are correlated with unexpected changes in stocks it is necessary to be able to decompose actual outcomes into their expected and unex-

1995 INVENTORIES AND THE BUSINESS CYCLE

FIGURE 2 Stock to Sales Ratio

Ratio

0.80

0.75

0.70

0.65

0.60

0.55

0.50

0.45

Non-farm

31

Ratio

0.80

0.75

0.70

0.65

0.60

0.55

0.50

Note: The non-farm stocks to sales ratio is from ABS 5206.0. The manufacturing stocks to sales ratio is from ABS 5629.0. Data are seasonally adjusted, at 1989190 prices.

pected components. In the absence of direct observations on expectations, a popular method of generating estimates of the predicted change is to estimate a time-series model. The miduals of this model are then treated as the unexpected change.

The justification for this approach is as follows. Suppose that both demand (D) and inventory investment (I) can be represented by simple ARMA( 1,l) processes:

(1)

(2) where E, and pr are serially uncorrelated error terms with zero mean and are independent of both D and 1. At time t-1. the expected value of demand at time t is Y D , - ~ + Thus, the fore-

D, = YDr-1 + E, + PEP1

4 = XI,-1 + H + @,-I

cast error. or the innovation in demand is E,. Sim- ilarly, the forecast error for inventory investment is p, If E, and p, are positively correlated then surprises in demand can be said to be typically associated with surprises in inventory investment of the same direction.

Typically, this approach suggests that unexpected changes in demand arc either uncorrelated, or positively correlated, with unexpected changes in inventories6 Below we argue that this approach has an important flaw and that survey based meas- urcs of expectations provide a potentially superior means of decomposing actual changes into their expected and unexpected components.

See Blanchard and Ascher (1989. p.16) and the working paper version of this paper.

32 ECONOMIC RECORD MARCH

TABLE 1 Variance of Sales and Output

Ratio of Comlation Variance of variance of Variances Between Deandcd Demndcd Y/S Using Log Log

sales output Changes in Changes in (S) ( Y) output and Salesand

Sales Inventories ~

Retail Trade 104 093 I29 019 1.24 I .60 0.02 Manufacturing 842 771 938 450 1.1 1 1.21 -0.07 Food & Beverages 47 236 48 625 1.03 1.19 0.10 Textiles 1401 2 062 1.47 :, 2.41 0.0 1 Clothing 6 691 7 755 1.16 1.30 0.35** Wood & Furniture 1 I 072 12 128 1.10 1.58 0.28* Paper & printing 7 708 7 981 I .04 1.15 0.06 Chemicals & Petrol 33 567 37 674 1.12 1.31 0.0 I Non-medlic Minerals 22 145 22 850 I .03 1.21 0.08 Basic Metal Roducts 31 434 33 166 I .06 1.28 0.10 Fabricated Metal products 24 863 26 855 I .08 1.26 0.10 T m p o n Equipment 36 91 1 45 001 I .22 I .52 -0.13 Other Machinery 36 834 42 144 1.14 1.26 0.15 Miscellaneous 8 130 9 183 1.13 I .35 0.04

Notes: 1. Retail mde d e s are from ABS 8501.0. Manufacturing sales (and its components) are from ABS 5629.0. These data are seasonally adjusted and arc at 198485 prices. Data on inventory investment are for finished goods stocks only and come from unpublished ABS data. They arc seasonally adjusted using the XI I package and deflated using the implied manufacturing stocks deflator. Ourput equals sales plus the change in stocks.

2. * (**) indicates that the comlation coefficient is statistically different from zero at the 5 ( I ) per cent level. The significance levels are calculated by regressing sales on inventories and using the Newey-West comction for the variancecovarimce matrix.

IV Demand, Costs and Inventories: Evidence from Survey Data

Since 1960 the quarterly ACCWestpac Survey of Industrial Trends has asked firms a number of questions concerning their actual and expected changes in production, stocks, orders. prices and costs etc. Specifically. the survey asks the following question for a range of variables: 'excluding nonnal seasonal changes, what has been your company's crpcrience over the part three months and what changes do you expect during the next three months?' Companies are asked to answer increase. decrease or remuin rhc s m . Unfortu- nately. the survey does not publish the responses of individual firms. It does, however, publish the percentage of firms that answer 'increase', 'decrease' and 'remain the same' to each question.

The survey consists of between 200 and 250 firms in the manufacturing industry. The sample of firms is basically constant, although f m are removed from the sample for repeated non response. Periodically, new firms are added to replace those f m s removed and to replace firms

that may have been taken-over or ceased to exist. The sample is designed so that the size distribution of surveyed firms reflects, as closely as possible. the actual size distribution of manufacturing firms in the economy.7 The sample period runs from the third quarter of 1960 to the third quarter of 1992.

By asking questions concerning actual and expected experience. the survey allows us to decompose actual changes in variables between expected and unexpected changes. For example, the unexpected change in demand at time t can be calculated by taking the difference between the percentage of firms that actually report an increase in new orders at time t and the percentage that, at time r-1, thought new orders would increase. Similar calculations can be performed to obtain unexpected changes in other variables.

In addition to using the difference between the

In the June 1992 survey, 36 per Cent of tirms had fewer than 101 employees, while 20 per cent had more than lo00 employees.

1995 INVENTORIES AND THE BUSINESS CYCLE 33

percentage of firms that experience and expect an increase in orders, stocks etc., we also use the difference in the percentage that actually report a decline in orders, stocks etc., and the percentage that expected a decline. Finally, we also use the net balance figures. The actual (expected) net balance is the difference in the percentage of firms that report (expect) an increase and the percentage that report (expect) a decrease. All three measures of the shocks should provide similar results.

One potential problem in using the survey responses to measure expected and unexpected changes arises from the fact that forecast errors will only be identified if demand moves in the opposite direction to that predicted. For example, if a firm expected that demand would increase by 5 per cent, but it actually increased by 20 per cent, the survey measure would not record an unexpected change for that firm. Certainly, if all firms faced the same conditions this would be a considerable problem. However, in practice, different firms have different expected and actual outcomes-at every point in time some firms will be expecting an increase in demand while others will be expecting a decline in demand. Given that a distribution exists concerning the size of expected changes, a general economy-wide improvement in expected demand growth should see more firms answer ‘increase’ to the question concerning expected changes in demand. If the distribution across firms of the expected size of the change in demand is uniform and, at least some firms are expecting an increase and others a decrease, the survey should provide accurate measures of expected and unexpected changes. While we know that not all firms agree about the expected direction of change, we know nothing about the shape of the distribution of expected changes. While this leads to caution in interpreting the results, we argue below that this approach offers an attractive alternative to time-series based techniques.

The unexpected changes (or shocks) to.both orders and the stocks of finished goods calculated using the ‘net balance’ figures are presented in Figure 3. There are two interesting points to note. First, the shocks to orders arc typically unfavourable shocks. That is, the number of firms that actually report an increase in orders is generally less than the number that predicted an increase in orders. Over the entire sample period. the average ‘shock‘ is -8.5 per cent. If the forecasts are unbiased this average shock should be insignificantly different from zero. A test of the unbiased hypoth-

esis is, however, overwhelmingly rejected.* Firms appear to be consistently excessively optimistic about their new orders. This optimism is also reflected in the fact that stocks of finished goods typically increase by more than was expected. The average expectation error is +7.9 per cent. Again, this is significantly different from zero.

More interesting for the question at hand is the apparent negative correlation between the shocks to demand and inventories. A number of episodes clearly stand out. The unexpected declines in orders in the second half of 1974, 1982 and 1990 were all associated with an unexpected increase in inventories. Similarly; in the 1960s unexpected increases in inventones occur when demand is unexpectedly low. In Table 2 we present the correlation coefficients between the forecast errors for orders and the forecast errors for inventories of both finished goods and raw materials. Corre- lations are calculated using the ‘up’, ‘down’ and ‘net balance’ figures. They are calculated using data from the entire period as well as data for each half of the sample.

Using data over the entire period we find a strong negative and statistically significant relationship between shocks to orders and shocks to inventories of final goods. Over the full sample period the correlation coefficient (using the net balance figures) is -0.40, which is significantly different from zero at conventional significance levels. In contrast, the correlation using the stocks of raw materials is small and is insignificantly different from zero. Similar results are obtained using both the ‘up’ and ‘down’ responses9 These results suggest that when demand for a firm’s product is unexpectedly low, its stocks of finished goods are unexpectedly high, but that there is no unexpected change in the firm’s stocks of raw materials. This indicates that when faced with an adverse shock to demand, firms do not reduce their production immediately but instead build up their stocks of finished goods. This is consistent with inventories of finished goods being used, in the first instance, as a buffer stock to demand shocks. Given that production and inventory accumulation initially continue in the face of the demand shock, it is hardly surprising that there is no immediate unexpected accumulation of raw materials stocks.

8 The t statistic for the test of the hypothesis that the

9 Given this similarity, we report results using only average shock equals zero is 6.7.

thc ‘net balance’ f i g m in subsequent tables.


-1 0

-20

-30

-40

F I C ~ R E 3

Unexpected Changes in Orders nnd Stocks of Finished Goods

Finished goods stocks Net

balance 30

20

10

0

-1 0

-20

-30

The results in Table 2 also suggest that there has been a change over time in the relationship between innovations in orders and innovations in inventories of finished goods. Over the first half of the sample period the correlation coefficient between demand forecast emrs and finished goods inventory forecast errors is -0.57, while over the second half of the sample period the coefficient is just -0.15. The hypothesis that the coefficients are the same can be rejected at the 5 per cent level.

These results stand in contrast to those obtained from generating unexpected changes from the residuals of ARlMA models. In part, the difference in results reflects the fact that the ARIMA models use only past values of the series and past error terms in generating expected changes. This information set is much smaller than that actually used by firms when forming their fortcasts. While the forecasts from the M A models may be

unbiased, the limited information set introduces a potentially important problem. Suppose that the innovations from the ARMA models (er and p,) each consist of two components and that each of these components has a zero mean. That is:

E r = +r + 'P r

PI = V I + c, (3)

(4)

and E(+) = E ( 9 ) = E(v) = E(<) = 0. Let the first component in each of the ARMA innovations represent the true innovation and the second component the error made by the econometrician through not being able to observe the complete information set available to firms. Assume that the errors induced by incomplete information are uncomlated with the true innovations (E(+q) = E(v<) = E(cpv) = E(K) = 0). Given this assump tion. the cornlation between the two ARMA innovations is given by:

1995 INVENTORIES AND 'ME BUSINESS CYCLE 35

4 E J . q = EN, + 'Pf)(VI + &,)I = E[$,v,l + 4cp,&,l. ( 5 )

TABLE 2 Correlarions Between Unexpected Changes in Orders

and Inventories

Sample Period

Inventories Of: 60:3-92:3 60:3-76:3 76:4-92:3

Finished Goods -0.30** -0.48** -0.03 -0.30** -0.49** -0.06 0 Down

0 Net balance -0.40** -0.57** -0.15 Raw Materials

0 Down 0 Net balance 0.04 -0.12 0.23

Notes: 1. (**) indicates that the comlation coefficient is smtistically different from zero at the 5 ( I ) per cent level.

2. The correlations for 'up' are the comlations between the forecast errors calculated using only the 'increase' responses for both orden and inventories. The 'down' correlations use only the 'decrease' responses.

UP

0.1 1 -0.01 0.27. 0.05 -0.17 0.29'

UP

TABLE 3 Correlations Benveen Expected Changes in Orders d Inventories (Net Balance Figures)

Sample Period

Inventories of 0.71 ** 0.58** 0.72"

Inventories of 0.82** 0.78'. 0.80.' Finished Goods

Raw Materials

Note: ** indicates that the correlation coefficient is statistically different from zero at the 1 per cent level.

The results from the survey data suggest that the true innovations are negatively correlated; that is. E($v) is less than zero. Thus if E(Ec() is approxi- mately zero, then E ( q Q must be positive. Given that the same unobservable information is likely to be used in forecasting both demand and inventories, it is not surprising that the errors induced by incomplete information are correlated. The fact that they are positively correlated suggests that when information not included in the ARIMA

model leads firms to expect demand to be low, this same information leads firms to expect that inventories will also be low. The discussion in Section I1 suggested such a positive correlation between expected changes in demand and inventories, provided that changes in demand had a rea- sonable degree of persistence and that marginal cost did not increase too quickly with output.

Table 3 presents the correlation coefficients between expected changes in demand and inventories of finished goods and raw materials. Each of the correlation coefficients in the table is positive and relatively large. In every case it is possible to reject the dull hypothesis that the correlation coefficient equals zero. Higher expected demand leads firms to expect both higher stocks of raw materials and higher stocks of finished goods. This suggests that production smoothing considerations are not of overriding importance and that the ARIMA models inade- quately distinguish between expected and unexpected changes.

We turn now to an examination of cost shocks. Ideally. we require some measure of shocks to real production costs per unit of output. While there is no direct measure of these shocks, the survey does allow the construction of a proxy. This proxy is constructed from two questions. The first question asks firms whether they expect an increase or decrease in their average costs per unit of output. These costs are in nominal terms. In an attempt to control for increases in the price level we use a second question which asks whether the firm expects their average selling price to increase or decrease. We use the difference between the net balance response to the cost question and the net balance response to the price question as a measure of expected changes in real costs. A similar calculation is performed using the data on actual experience. Subtracting the expected experience from the actual experience provides us with our proxy for cost shocks.'O

In Table 4 we report the correlations between these cost shocks and the shocks to inventories of

'OThis proxy for real cost shocks is less than perfect. For example, if a firm experiences no change in its cost of production but is able to increase its output price (with no change in the general price level) our measure would record a reduction in real costs. However, this would be incomct as our measure would simply be picking up an increase in margins for the individual firm. This problem is less severe when we aggregate over a large number of firms as not all firms can increase. their prices relative to the general price level.


TABLE 4 The Relationship Between Inventories and Costs

(Net Balance Figures)

Sample Period ~ ~~~~

66:3-92:3 66~3-16~3 16~4-92:3

Cornlation Between Unexpected Inventories and Unexpected Costs Inventories of:

Finished Goods 0.12 0.10 0.27' Raw Materials -0.15 -0.19 -0.07

Cornlation Between Expected Inventories and Expected Costs

Finished Goods Raw Materials

-0.32" - 0.29' '

-0.34 -0.16

-0.58" -0.58.'

~ ~~ ~ ~~

Note. *(**) indicates that the correlation coefficient is statistically different from zero at the 5 ( I ) per cent level.

both final goods and raw materials.lI We also report the correlations between our proxy for expected changes in real costs and expected changes in stocks.

For the sample period as a whole, the correlations between the cost shocks and the inventory shocks are small and insignificantly different from zero.12 In contrast, over the second half of the sample period the correlation between cost shocks and shocks to inventories of finished goods is positive and statistically significant. However, this correlation is opposite in sign to that predicted by cost-shock based inventory models; it suggests that when costs are unexpectedly high, inventories are also unexpectedly high.

In contrast, the negative comlations in the second half of the table provide at least some support for the cost-shock theories. This is partic- ularly the case over the second half of the sample period. When costs are expected to increase. firms expect both stocks of finished goods and raw materials to fall.

In estimating and presenting the above correlations it has been implicitly assumed that cost shocks and demand shocks are uncomlated. This assumption may not be valid. When demand is weak, the costs of production may fall and when

1 1 The questions conceming costs and prim wen introduced in Jum 1966. Consequently. dK sample period used to calculate cmlations betwan invmto- ria and costs tuns from June 1966 to September 1992.

l2 One possible explanation for thc zclo comlation is a lag in adjusting productioo in response to a change in costs.

demand is strong production costs may rise. If demand and cost factors are related, the above correlations may not reflect underlying independent influences. In an attempt to isolate the independent effects of demand and supply side factors we estimate two equations. The first equation explains shocks to inventories in terms of the demand and cost shocks. The second explains expected changes in inventories in terms of expected changes in demand and expected changes in costs.

Rather than imposing the constraint that the inventory response is symmetric with respect to increases and decreases in demand, we allow the response to vary depending upon whether the firm is expecting a favourable or an unfavourable change in demand. Separate estimates for the two cases are obtained by defining two dummy variables. The first dummy (DI) takes a value of one if the expected change in demand is greater than the mean expected change.13 The second dummy (M) takes a value of one if it is less than the mean. The expected demand series is then multi- plied by these two dummy variables to create two 'expected demand' regressors. A similar procedure is used to allow for asymmetries in the response to expected changes in costs. The dummy variable 0 3 (W) takes a value of one if the increase in the 'real' average cost is expected to be grcater (less) than the average expected change. Analogous dummy variables are also

13 Separate means arc calculated for each of the three sample periods.

1995 INVENTORIES AND THE BUSINESS CYCLE 37

TABLE 5 Explaining Unexpected Changes in Inventories

Inventories of Finished Goods Inventories of Raw Materials

66~3-92~3 66:3-76:3 76~4-92~3 6613-92~3 6613-76~3 76:4-92:3

Constant

Demand Shock*Dl (favourable) Demand Shock*D2 (unfavourable) Cost Shock*D3 ( unfavourable) Cost Shock*W (favourable) R?

852 (6.18) - 056 (2.25)

-0.12 (1.35)

-0.03 (0.19)

- 0.24 (0.73) 8.18

13.63 (3.66) - 0.96 (8.67)

-0.14 (0.99) - 0.45 ( 1.72) - 1.06 (3.47) 052

7.19 (4.75) 0.14

(1.06) -0.08 (0.82) 0.24

(2.22) 0.29

(0.73) 0.03

10.90 (5.34)

-0.14 (0.89) 0.11

(1.14) -0.16 (1.14)

-034 (0.98) 0.0 I

15.22 10.2a (2.84) (5.01)

-039 0.27 (2.5 I ) (2.55) 0.11 0.13

( 1 .00) (1.10) -0.48 0.02 ( 1.30) (0.14)

-0.95 -0.28 ( I .97) (0.77) 0.10 0.01

H i p , = p, @-value) 0.07 0.00 0.2 1 0.15 0.01 0.45 H,:$, = $, @-value) 0.49 0.00 0.90 0.57 0.02 0.40

Note: 7-statistics appear in parentheses below coefficient estimates. Standard errors have been calculated using the Newey-West procedure with three lags.

defined to allow for asymmetric response to unexpected changes in demand and costs.

The estimation results for the equations explaining shocks to inventories are presented in Table 5 , while the results for the equation explaining expected changes in inventories are presented in Table 6. Using the full sample period, the results confirm that demand shocks have had a statistically significant effect on inventories of finished goods. The effect is significantly stronger when demand is unexpectedly high than when it is unexpectedly low. The coefficients on the cost shock variables are insignificantly different from zero. Comparing the results for the two sub- samples, we find significant differences in the estimates and in the ability of the equation to explain shocks to inventories of finished goods. In the first sub-sample the coefficient on the favourable demand shock variable is highly significant while in the second sub-sample it is insignificantly different from zero. In the first sub-sample both cost shock variables have the correct sign, but in the second sub-sample they have the incorrect sign. Perhaps, more importantly, over the first period the R2 is 0.52; this falls to 0.03 over the

second half. This suggests that unexpected changes in demand and costs have come to play a significantly smaller role in the inventory cycle.

The equation explaining unexpected changes in the stocks of raw materials has very low explan- atory power. Over the full sample period, both the demand and cost shock variables are insignificantly different from zero. In contrast. when the sample is split in two, the favourable demand shock is significant in both sub-samples. The sign is, however, not consistent across the two periods.

The equations explaining expected changes in stocks perform considerably better. Over the full sample period, both the expected demand variables have positive and statistically significant coefficients in both the stocks of raw materials and stocks of finished goods equations. That is, when firms expect demand to change they also expect their stocks of raw materials and finished goods to change in the same direction. In both equations the cost shock variable is insignificant when using the full sample period. For finished goods the two sample periods

again show different results. Over the first sub- sample, the coefficients on favourable and unfa-


TABLE 6 N b i n i n g Erpccted changes in Invtntonts

Expected A expected A erpcced A

Inventories of Finished Goods Inventories of Raw Materials

Constant

Expected Demand*DI (favourable)

Expected Demd*D2 (un favourable) Expected Change in Costs*D3 (unfavourable) Expected Change in Costs*D4 (favourable) &

66:3-92:3 66:3-76:3 76:4-92:3 663-92:3 66:3-76:3 76:4-92:3

- 8.m -6.60 -733 - 1269 -11.22 - 1 1 3 8 (4.25) ( I .79) (3.94) (5.94) ( 1.77) (6.38) 021 021 0.05 0.35 0.37 020

(3.09) (3.38) (0.45) (4.96) (4.55) ( I .83) 034 0.07 0.45 0.47. 0.29 052

(3.01) (0.97) (5.71) (4.88) (3.32) (6.25) -0.14 -0.18 -0.28 - 0.07 -0.64 -0.25 ( 1.08) ( I .39) (3.28) (0.45) (0.13) (2.10)

-0.26 -0.11 -0.26 -0.22 -0.16 -0.16 (1.13) (0.52) (1.45) (0.93) (0.44) (0.84) 0.47 0.34 0.60 0.63 057 0.67

/f,:p, = @-value) 0.39 0.25 0.0 I 0.39 0.59 0.04 H$B, = 8. @-value) 0.37 0.55 0.86 0.23 0.37 0.47

~~ ~ ~ ~~

Note: T-statistics appear in parenlhcses below coefficient estimates. Standard emrs have been calculated using the Newey-West procedure with three lags.

vourable changes in demand arc not significantly different from one another. In contrast, over the second sub-sample. the coefficient on unfavourable changes in demand is significantly greater than that on favourable changes. O v a this second sample period, expected declines in demand have caused large expected declines in inventories. while expected increases in demand have caused negligible increases in inventories of finished goods. In addition, unfavourable expected changes in costs have had a significant effect on expected inventories of the predicted sign. The sign on the favourable shock is also that predicted by the theory, although it is insignificantly different from zero.

The results in Tables 5 and 6 suggest that changes in stocks of finished goods have increasingly become driven by expected rather than unex- peaed changes. The ability of demand and cost shocks to explain unexpected changes in inventories is much lower over the second sub-sample than it is over the first sub-sample. In contrast, the ability of expected changes in demand and costs to explain expected changes in inventories is considerably higher over the second subsample. For

finished goods the R2 increases from 0.34 to 0.60 while for raw materials it increases from 0.57 to 0.67.14

V Summary Md Conclusions In this paper we examine the interaction of the

inventory cycle and the business cycle. In particu- lar, we discuss the influences driving the inventory cycle and the impact that inventory investment has on the business cycle. These issues are examined with the aid of both macroeconomic data and data from surveys of individual firms.

In contrast to an increasing volume of literature supporting the view that changes in inventories are driven by shocks to the cost of production, we find that demand factors play an important role. Using survey data rather than macroeconomic data, we show that unexpected changes in inventories have typically been associated with unexpected changes

14 Tbe above results were qualitatively unchanged when we included a time trcnd in the regression and when we split the sample period.at June 1982 (the point at which the stocks to sales ratio begins its trend decline).

I995 LNVEMoRlES AND THE BUSINESS CYCLE 39

in demand of the opposite sign. In contrast, when demand is expected to change, inventones are expected to change in the same direction as the change in demand. This suggests that shocks to demand are viewed as having a considerable degree of persistence andor that production smoothing is relatively unimportant. While the results do not rule out a role for cost shocks in influencing the inventory cycle, the survey data provide considerably stronger evidence in support of a central role for demand variables than is typically found using macroeconomic data.

Perhaps more importantly. the results also show a significant change in the behaviour of inventories over the past decade. In the business cycles in the 1960s and 1970s, unexpected falls in demand led to a substantial increase in inventories. This was followed by a cut-back in production as firms attempted to wind back the excessively high level of inventories. As a result, the inventory cycle amplified the business cycle.

More recently, unexpected changes in demand have come to play a much smaller role in explaining the behaviour of inventories. The notion that production smoothing is very important and that firms use inventories to buffer production from changes in sales does not appear to be a good description of reality. Increasingly, inventories move positively in line with expected and actual changes in demand. In part, this reflects increasingly sophisticated inventory management techniques and greater production flexibility. The fact that firms can more effectively prevent unintended inventory accumulation in the face of adverse shocks. reduces the need for firms to subsequently cut back production to iun down inventories. The reduction in the stocks to sales ratio has also made inventories less sensitive to demand shocks. The result of these changes is an inventory cycle which has a smaller amplitude and a reduced impact on the output cycle.

REFERENCES

Brason, Richard (1993). 'Tests of Production Smooth- ing in Selected Japanese Industries', Journal of Monefary Economics 31, 38 1-400.

Blanchud. Oliver (1983). 'The Production and Inven- tory Behaviour of the American Automobile Indus- try', Journal of Political Economy XCI, 365400.

- and Stanley Fischer (1989). Lccrures in Macro- economics, MIT Press, Cambridge, MA.

Blinder, Alan (1986). 'Can the Production Smoothing Model of Inventory Behaviour be Saved?', The Quarterly Journal of Economics CI. 43 1-54.

- and Louis Maccini (1991). 'Taking Stock: A Crit- ical Assessment of Recent Research on Invento- ries', Journal of Economic Perspectives 5, 73-96.

Flood, Darren and Lowe, Philip (1993). 'Inventories and the Business Cycle', Reserve Bank of Australia. Research Discussion Paper 9306.

Kahn, James (1992). 'Why is Production More Vol- atile than Sales? Theory and Evidence on the Stock- out-Avoidance Motive for Inventory Holding', Thr Quartedy Journal of Economics CVII. 48 1-5 10.

Metzler. Lloyd (1941). 'The Nature and Stability of Inventory Cycles', Review of Economics and S i n - tisrics XXIII. 1 13-29.

Summers, Lawrence (198 I). 'Comment', Erookings Papers on Economic Activify 2. 5 13- 17.

West, Kenneth (1988). 'Evidence from Seven Coun- tries on Whether Inventories Smooth Aggregate Output', Discussion Papers in Economics No. 140, Woodrow Wilson School. Princeton University.

- (1989). 'The Sources of Fluctuations in Aggregate Inventories and GNP'. NBER Working Paper No. 2992.

inventories and the business cycle

Documents