Price stabilisation and agricultural supply

PRICE STABILIZATION AND AGRICULTURAL SUPPLYN Jean-Marc Boussard Françoise Gérard When prices are falling, no matter when, no matter where, farmers always demand "price support" - in any case, state intervention - . In developed countries, such support has usually been granted, and on a massive scale since the crisis of 1929 and the literature it produced*. The damages caused by the crisis were considerable, whilst the actual stabilization cost was not, so that operations were carried out without giving rise to much discussions among the economists2. . As far as developing countries are concerned, thought was carried much further. General price stabilization systems were considered worldwide. Often, they failed to be implemented (and perhaps quite rightly, in view of the research conducted). A certain number of countries pursued on their own behalf an active domestic agricultural price regulation policy. For example, this was seen in Thailand from 1960 to 1985 (rice) and in the Ivory Coast from 1958 to 1989 (cocoa). However, as a rule, international markets remain remarkably free and fluctuating, as do many home markets. Economic doctrine on the subject derives to a great extent from work done on a large scale by the World Bank and summarized in the positively monumental work by Newbery and Stiglitz (1981)3 . Their chief conclusion is that price stabilization, a costly and difficult operation, is N OTES * This study was financed in part by the F.A.O. 1. Depicted by, among others, John Steinbeck in "The Grapes of Wrath". 2. However, the work of Ezekiel (1938) on the Cobweb, followed by that of Massel (1969), Oi (1961), Waugh (1944) and many others, was partially caused by these events. . In the same way, it is also worth consulting Schmitz (1984), Scandizzo, Hazell and 3 Anderson (1985). 9 probably of less use to the producer than to the consumer, and that there are more effective ways to support farm income4. . As a result, they do not recommend it. Without wishing to question the conclusion drawn by Newbery and Stiglitz, which has been much simplified here for reasons of space, one might ask if their analysis could be completed by some additional considerations that might add a new dimension to their findings. First, it will be shown that price stabilization is not simply a matter of dividing a potential surplus among producers, consumers and possibly taxpayers. It is something that is essential for determining the volume of production and can be regarded as another aspect of technical progress. Secondly, and on a practical basis, some statistical evidence to support the relevance of this approach will be provided. THEORETICAL EFFECTS OF STABILIZING SUPPLY It is a well-known fact that a decision maker, facing a risky situation, does not maximize the mathematical expectation of his gain, but rather something more subtle, known technically as the Certainty Equivalent of the gain. The Certainty Equivalent of a random quantity is the "certain" value that has to be given to this quantity if it is to lead to the same decisions as those found in a random situation5. . If I don't care whether someone gives me four 100- franc notes right now or a lottery ticket that gives me one chance in two of winning 1000 francs tonight, then for me 400 francs is the certainty equivalent of one chance in two of winning 1000 francs. The fact that the sum of 400 francs is lower than my lottery expectation, i.e. 500 francs, shows that I am aware of the risk and that I would not be tempted by a lottery ticket without a "risk premium" of 100 francs at least. This sort of "aversion to risk" is a common phenomenon, particularly among farmers, and causes them to produce less than they would if there were no risk (see Figure 1). Figure 1 [demande] Demand [offre] Supply [demande dérivée] Surrogate Demand On a classic graph (P,Q) we show the marginal cost (a value known with certainty) and the demand. Optimum equilibrium on the market is reached at point A, when marginal cost is * * equal to demand, for quantity Q and price P . This is normally the result of free competition. However, if we assume (as is only practical) that the farmer is unsure of the price, then he will not try to intersect the supply curve with the price he hopes to get, but rather with the certainty equivalent of this price. In presence of risk aversion, this certainty equivalent is 4. Along the same lines, Newbery (1989) advocates stabilizing the prices of foodstuffs because of the disastrous effect of fluctuations on consumers, especially the poorer consumer. 5. There exist numerous theoretical references on this subject. Suffice it to quote here the well-known article by Allais (1953), who very rightly criticizes classic theory on the question. 9 lower than the average price, so that equilibrium corresponds, on point B, to the intersection of the marginal cost with a surrogate demand curve situated constantly below the real curve. * * Hence the quantity produced is Q < Q and the price is P > P . The average profit earned by the farmer as a result of this operation is a risk premium, which increases proportionally to the risk. From this point of view, Newbery and Stiglitz were completely right to emphasize that price stabilization worked more to the advantage of the consumer than to the producer. This approach does, however, obscure a major aspect of the reality of the situation, namely that as a general rule farmers are not equipped to make the same sort of profits as do insurance companies or banks. And the mechanism whereby such profits are obtained has an unpleasant side to it. If it may be true that some reductions in supply result from explicit decisions taken by farmers (perhaps, for example, rather than use all their liquidities for buying inputs, they invest them as unproductive assets - the savings bank in France, undernourished livestock in Africa), most of the time they occur because, at certain times, falling prices lead to bankruptcy. Even if, in the farming business², firms are more or less protected from bankruptcy by their ability to defer many payments, on genuinely free markets, the situation only intensifies the problem, because continuying production makes prices falling lower. In addition, demand rigidity does not make the necessary adjustment easier6. W . Finally, what happens7. is that some particularly well-placed farmers make a fortune when prices return to famine levels after a fall, whilst other farmers find themselves impoverished at the very time that the inevitable price increases make the situation extremely difficult for the consumer. When prices are stabilized, on the other hand, farmers can expand their activities .They show no profits, stricto sensu, but because they can employ a greater quantity of factors, they get incomes in excess of that which they would obtain in a free market. Thus it is not difficult to understand why farmers are basically in favour of stabilization. Stabilization is not without its problems, even if Newbery and Stiglitz do not appear to have been aware of the real difficulties involved in effective stabilization of world prices. They regard any such stabilization with scepticism, because of practical storage problems, storage being both costly and very possibly inadequate. More especially, they point out that if the volume of stock results from a random walk, created by a Gaussian law of probability, i.e. the volume of stock in year n equals that of year n - 1, increased or decreased by a random quantity assumed to be normal, then there exists a non-zero probability that the stock will exceed any predermined level after a longer or shorter time (Newbery and Stiglitz, 1981. p. 6. When demand is rigid, a small variation in supply causes large price variations. As a consequence, a slight overproduction may cause prices falling very low. It has been said (Especially, by Newbery and Stiglitz) that in such cases, because prices and quantity variations are of opposite sign, incomes should not be too dramatically affected. But this is true only if the production achieved by each individual farmer is totally correlated with overall market supply. Such a situation seem to be somewhat rare . 7. These ideas would deserve a formal exposition that lack of place prevents to present here. It should be noticed that we are looking less at a risk exogenous to the farm sector, caused, for example by sunspots or the weather, and more at an endogenous risk arising from mechanisms of the market itself. A discussion of this particular aspect would be outside the scope of the study. See Boussard and Gérard (1992a). 9 34). However, in practice, if harvest deviations from the mean were really Gaussian, exceeding storage capacity would be such a rare event as to be excluded for all practical purposes, and the Newbery and Stiglitz argument would not hold. In fact, it holds, but for a different reason: the very existence of stabilization could cause a situation in which storage capacity was exceeded. If, as is probably the case, the agricultural production function is homogeneous and of degree one8. , the slope of the marginal cost curve is no longer positive. It is completely flat, horizontal and equal to average cost (see Figure 1), which is itself constant. Under these conditions, and as long as the demand remains a decreasing function of price, there is always an optimum level of production where cost equal price. But if a successful price stabilization system is introduced, the demand curve would become parallel to the axes. From then on, there would be no limits to the increase of supply, thus defeating any attempt to put a limit to stabilization payements. This can help to explain the grain surplus in Europe and the United States as well as the cocoa surplus in the Ivory Coast following two or three decades of guaranteed prices. All this obviously has to be interpreted in terms of dynamics rather than statics. Production growth is never instantaneous, because in the short term there are always fixed factors which make marginal cost increasing and in excess of average cost. But in the long term, the more farmers earn, save and reinvest to escape from fixed factors, the more the marginal cost curve tends to parallel the Q axis, and the more production increases - until, and inevitably, the price support policy becomes impossible, i.e. too costly, as happened in the Ivory Coast and as is currently happening in the EEC. This is somewhat speculative. The question that needs to be asked now is how to find empirical proof of the hypotheses that have been put forward, and we shall be examining this in the second part of our study. 8. If it is possible and profitable to produce 100 tonnes of wheat with ten hectares of land, one man and one plough, it is equally possible and profitable to produce 200 tonnes on twenty hectares, using two men and two ploughs; or indeed a thousand million tonnes on a hundred million hectares, using 10 million men and ten million ploughs. And so on. 9 TESTING THE HYPOTHESIS METHODOLOGICAL PROBLEMS The effects of risk on agricultural supply have formed the subject of two types of contradictory empirical studies. On the one hand, at the end of the fifties, numerous authors produced models based on mathematical programming and which treated the farmer's choice as though it were some sort of portfolio on the stock exchange. Thus Freund (1956) showed that such a model is far more representative than an "ordinary" linear program of the actual behaviour of farmers in North Carolina. This pioneering work has since been reproduced a number of times, and always with the same degree of success. The standard linear programs produce solutions in which the crops most at risk are always produced in greater quantities, because they always carry a higher risk premium and are therefore, on average, the most "profitable". When the utility function comprises a penalty for risk, these activities either disappear or are relegated to a less important role in the cropping system. Models of this type, which explicitly associate financial and production decisions do give, from this point of view, better results, that is, optimal plans closer to thoses actually observed (Boussard, 1980; Just and Zilberman, 1986). As a result, it may be concluded that risk plays a major role in the decisions taken by farmers. Oddly enough, as a rule, these findings are not confirmed by macroeconomic statistical inference methods. The most natural idea, from this point of view, would consist of working out the following model: (1) [formule] where q is the quantity produced and p the price of a farm product while is a random disturbance. It should be shown that it is possible to improve the predictions provided by a model of this type by specifying it differently: (2) [formule] where r is a price variability index9 . Unfortunately, it is extremely difficult to specify correctly a model such as (1) above. In such an endeavour, one is rapidly faced with almost all the most irritating difficulties in econometrics, from identification problems resulting from the fact that supply cannot be estimated independantly of demand, to the delicate issue of finding the right submodel for expectiations, since it is obvious that the "price" used in equation (1) cannot be the actual price at which the product is sold, which is unknown when decisions concerning production are taken. There is also the impossibility of allowing for variations in all the other prices likely to affect the farmer's decisions. For these reasons, and as was pointed out some years ago by an expert on the subject (Nerlove, 1979), the numerous attempts made in this direction over the last thirty years and 9 . In reality, this is the price risk, i.e. the surprise effect associated with a given price or, in other words, the difference between actual and expected prices that we hope to measure. However, any definition of such a variable is delicate; and there is no consensus as to how anticipations are created. 9 more have never provided really convincing results, even when the authors profess to the contrary10. . If in fact we cannot use a suitable relationship of type (1), then it is extremely difficult to go to type (2) and, as a result, to make a comparative evaluation of the two models. To that must be added another classical, but very real, difficulties, namely the lack of data. If we wished to put forward the propositions above with a minimum of generalization, we would need to ensure reasonable "coverage" of the majority of agricultural products in the world. The only sources available in this respect are those of the FAO, which provides for each product j, for each country k and for each year t the average price, the area and the relevant production11 . This gives a data base of about 2000 series, for 34 products in 80 countries, but it is obvious that such data are not sufficient to work out sophisticated models based on the evaluation of CES or Translog production functions parameters. Furthermore, it is reasonable to assume that these series are not entirely reliable and that measurement errors are to be be expected. That is why it was decided to leave out the comparison of models (1) and (2), and to rely on a different approach . METHODOLOGICAL CHOICES A large part of the difficulties just mentionned stem from the fact that the models to be estimated are , so speaking, too exact. It is not last year's price, or the price of two years ago that decides the farmer to plant this year. It is rather a somehow qualitative notion of "price" in general which plays a role in his decisions. "The price is good", or "bad." The crop is "sure" or "at risk". That is the language the farmers use when questioned, rather than forecasts implicit in most econometric models on the lines of "The price will be x in the third quarter of next year..." Thus one is induced to establish a qualitative rather than a quantitative classification of the price and quantity series provided by the FAO. On that basis, it becomes possible to adopt a "variance analysis" type of approach12. .It will be ascertained whether production associated with "stable" series is greater than that associated with "unstable" series. As mentioned above, the effects of price instability should be analyzed in terms of dynamics. The theory does not predict that stabilization immediately leads to infinite production, only that indefinitely prolonged stabilization13 leads to a 10. But see Antonovitz and Green (1990) or Berni et al. (1988), or again Trail (1978). . It should also be noted that not all crops and not all countries are covered. Analogous data 11 exist at the USDA but they cover an even more limited field. 12. Variance analysis offers the possibility of testing the effect of qualitative variables on a measurable magnitude. Cases are grouped together according to sets defined by the modalities of qualitative variables. The mean and variance values of the measurable variable in each set are computed. Tests are then conducted to ascertain whether, given variances, the differences in mean between sets are significantly different. . Needless to say, we are here talking about any transfer of risk between the producer and 13 whatever body is regulating the market. 9 limitless increase in production, and this leads one to consider not production in itself but rather its rate of increase. This line of argument means that we have to determine whether there are significantly higher increases in production in countries and for "stable price" products than in countries and for "unstable" products. It remains to define exactly what one means by "stable" and "unstable", and by "rate of growth of production". We shall use the following indices: j: products, j = 1...J k: countries, k = 1...K t: years, t = 1...T Jk represents the number of products present in country k. Pjkt represents the price of product j in country k in year t, and Yjkt the corresponding production. 9 From there, we can define: - qjkt, the instant crop price growth rate, i.e. [formule] - ajkt, the instant crop price variability: [formule] - the order p moving average crop price increase rate :14. [formule] - the country agricultural price index increase rate : [formule] (In this last formula, it would probably have been preferable to weight each data item by the weight of the product in the country's total production, but this value is not available.) - the instant crop production growth rate: [formule] - the order p moving average crop production growth rate [formule] - the average country crop production growth rate [The arithmetic mean of instant production growth rates] These concepts will be usefull in defining "stable" or "unstable" series. Conjunctural stability 14. Clearly, this measurement, as a price risk approximation, is far from perfect. However, in the absence of a detailed study by region on the creation of anticipations and price policies, it was difficult to do much better. Another problem associated with this definition is posed by the existence of a tendency (e.g. inflationary) which could be incorporated into expectations. This sort of scenario would question the relevance of the indicator used. A systematic graphical examination of the price series revealed no sign of strong tendencies. It must be noticed that prices series used in this study are deflated by the "agricultural GNP deflators " of the World Tables. 9 It can be decided that data item kjt is "stable" if vjkt < s; it is "unstable" if vjkt > u; and "doubtful" if s vjkt u, where s and u represent more or less strict thresholds . This method has the advantage of providing numerous cases for variance analysis. Additionally, these cases are time dependant (hence the term "conjunctural"), thus allowing for tacking into account possible policy changes during the time of observation. It must be noted here that vjkt values are relative average variations expressed as a percentage. Thus s = 0.20 means that we are considering as stable a case (that is, one crop, one year, one country ) such that the absolute value of the order p moving average of growth rate does not exceed 20% Structural Stability It is decided that a series ( that is, one crop in one country) is: - stable, if Freq {......................} ; - unstable, if Freq {................... } ; - doubtful if none of these relations is true. This concept only provides a limited number of cases; but it is probably true to say that it corresponds better to the basic idea underlying the steps to be taken. Definitions of "stability" and "instability" are necessarily somewhat arbitrary, as these terms each depend on two or four "threshold parameters" that the analyst is in some ways free to choose as he deems appropriate. In order to reduce the arbitrary nature of these choices, many runs of variance analysis have been performed with a large spectrum of values for these parameters. We shall see that from this standpoint, the results are quite consistent, and that the effects we wished to reveal are simply more strongly contrasted when the criteria governing the selection of "stable" or "unstable" cases are more strict. Results Table 1 shows the results of a set of one way variance analysis used to test the effect of structural stability. That is to say, average production growth rate ckj 's have been examined in order to see if it is significantly smaller for "unstable" cases. Tests have been performed for different frequency threshold values defining stability, with the set of moving averages calculated over 5 years. Table 2 shows the same type of results for conjuntural stability. Here, one case corresponds to a group of 5 years, during which time it is estimated that the price of a product in a given country has been "stable" or "unstable" on the basis of the indicated thresholds . Table 1 shows, quite evidently, the effect of stability or instability on the rate of growth. It also shows that this effect is only felt when the criteria used to define "stable" and "unstable" situations are strict. The F-test rejects the null hypothesis of "no influence of stability on average production growth rates" only for the few stable 100 series, for which more than 80% of the year-to-year differences in prices are less than 10% of the mean. It is, however, only natural that the phenomenon only appears here in "well-defined" cases. During the 20 or 25 9 years of these series, there may well have been periods of either stability or instability for the same product in the same country, which automatically gives rise to "doubtful" situations. This source of error should be eliminated in Table 2. Here, analysis is by overlapping 5-year periods. As might be expected, tests are by and large significant and growth rate deviations considerable. When the definition of "instability" becomes stricter, deviations are more marked and the tests are even more significant. 9  Table 1 Variance Analysis of Average Crop Production Growth Rates (Cjk), according to the criterion of structural "stability/instability" Higher Stability Requirement Higher Instability Requirement Value Value No. of cases In bold: significant results at a threshold of 2.5%. 1) Each box in the table corresponds to a variance analysis, for different definitions of structural stability. A series is classed as stable if Freq {.....................} > a15 , as unstable if Freq {..........} For different values of .......... and ..............., the table shows: - the percentage difference between the average production growth rates of the "stable" and "unstable" series; - the Fisher "F" used to test the homogeneity of variances between residuals , - the second element of the pair of degrees of freedom in the F test (the number of cases minus 2, the first element always being 1 by construction). The surprising phenomenon is that increasing the strictness of the definition of stability produces an inverse effect. The more the definition is tightened, by reducing the threshold beyond which a data item is considered to be "stable", the smaller the difference of the growth rate with the "unstable" series. No satisfactory explanation for this very considerable phenomenon has been found. Table 2 Variance Analysis of 5 Years Moving Average Crop Production Growth Rates, (Cjkt [5]), according to the criterion : "conjunctural stability/instability" Higher Stability Requirement Higher Instability Requirement u Value No. of cases s Value 15 . Here, the threshold a is given the value of 0.1. Other values of 0.25, 0.5 and 0.75 were attempted. With such values, the experimental design becomes very unbalanced, with the number of unstable series very low after a threshold of 0.25. 9 In bold: F-test significant at a threshold of 2.5%. 1) Each box in the table corresponds to a variance analysis, for different definitions of situational stability and instability. A data item, Ckjt [5] is considered "stable" if vkjt [5] s and "unstable" if vkjt [5] u. The table shows: - the percentage difference between average Ckjt [5] values of "stable" series and those of "unstable" series; - the Fisher "F" used to test the homogeneity of variances between residuals and the original data population. - the second element of the pair of degrees of freedom in the test (the number of data items less 2, the first one always being 1). Overall, these results confirm the theory we started with. But we might ask if the choice of a 5-year period for moving averages is too arbitrary. Five years does very probably correspond to a "normal" time span as far as expectations are concerned16. . But what would happen if we were to choose another type of moving averages? This has in fact been done, with very similar results. Thirty variance analyses have been carried out on the same model as the 25 shown in Table 2, with vjkt and cjkt values calculated on the basis of variable increment moving averages. The results obtained were fairly similar, in principle, to those shown in Table 2 and which we have attempted to summarize in Table 3. This last table only shows the proportion of significant F-tests of the 30 included in each table. Table 3 Effects Associated with the Length of the Moving Average Period. Proportion of Significant F-Tests for Variance Analyses on the Model of Table 2 Performed for 8 Moving Average Periods Moving Average period Frequency of Significant F-Tests at a 5% Threshold The table shows, obviously enough, that the risk effects are best captured through period of analysis lasting 4 to 10 years, which may help to understand the lack of success of studies which aimed at evaluating models of this type on a yearly basis. One might ask whether the same sort of analysis could be applied to prices. Is the production growth rate higher when prices can be regarded as "high" over a fairly long period? The problem there would be to define the meaning of "high". This would have required a comparison either with the price of the same product in other countries (and then we would have the exchange rate problem) or with other agricultural prices and the prices of inputs. We did not have input prices and any comparison with other prices would have been difficult. That is why we have asked a slightly different question and, in point of fact, one that is less significant from the point of view of economic theory. Is the rate of increase of production higher when the rhythm of price increase can be described as high? That question could be answered using the same method as before, by defining price increase thresholds beyond 16. "Normal time span" should be taken to mean the number of periods elapsed and commonly taken into account by an agent to give him some idea of future prices. 9 which increases are classed as "low", "average" or "high". Indeed, to eliminate the effect of general farm price increase policies, a comparison was made between hjkt and hkt values (average for the country being considered), cf. Table 4. The results17. are not significant. The average 5 years moving average of crop production growth rate are not very different when 5 years moving average of crop price growth rates are "high" or "low ", unless, however, the criteria used to define "high" and "low" rates of increase are very strict (although then the number of cases is considerably reduced). Similarly, it is tempting to try a two ways variance analysis, in an attempt to identify the Table 4 Average Rate of Production Increase (Cjk [5]) and Number of Cases according to the criteria of "stability/instability" and "high/low18. price increase" Price Price Increase Prod. Increase (%) Number of cases Stable High Low Unstable High Low respective effect of the means and of the variability of prices. The difficulty here stems from the need to "balance" the experimental basis (i.e. to have about he same number of cases in each situations). Whatever method used to solve it, this test (Table 4) confirms previous results. The criterion of "variability" leads to significant results. The criterion of "price increase" does not. Even more surprisingly, when variability is "high", it can be seen that the average rate of production increase is somewhat lower when price increases are "high" than when they are "low". This result is rather difficult to interpret, because a low rate of production growth when price increases are "high" may arise from the fact that "we started too low..." But it does make one wonder about the "automaticity" of the connection between an increase in price and an increase in production. At the same time, a negative result of this sort goes some way to reinforcing the validity of our conclusions. With "good" results for average prices (that is high rates of production growth being associated with high rates of price increase) , there is always the suspicion that "good results" for variability (that is, high rates of production growth being associated with 17. Described in detail by Boussard and Gérard (1992b). 18. One case (one set of values for j, k, and t) is "stable" if vjkt < s; "unstable" if vjkt > u. The corresponding price increase is "high" if hjkt > x.hkt, ,"low", if hjkt < y.hkt. 9 low price variability) may be due to the correlation that necessarily exists between "high variability" and "high rates of price increase", simply because one is obtained on the basis of the absolute value of the other. Thus it can be seen that the effect of taking the absolute value of increases in price is to increase rather than decrease the explanatory power of price variations for changes in the level of production. That might appear paradoxical, but it is in accordance with the pure theory expounded above. A major upward change in price conveys in fact an ambiguous message. On the one hand, if the change is maintained, the profitability of the crop in question increases. But on the other hand, if the change is not maintained, that only means that the crop represents a risk. Thus the farmer cannot interpret the message out of context, and therein lies the whole problem of expectations and the dynamics of agricultural supply. 9 CONCLUSIONS Whatever the approach, the "price stability/instability" dichotomy would appear to be more effective in explaining increases in the volume of the supply of farm products than the "high/low price increase" dichotomy. Despite the fact that the "risk factor" has long been recognized as a major determinant of agricultural supply, very little statistical confirmation of this assertion has been published until now. Thus the present study fills a gap in the relevant literature, even if "price variability" is not to be confused with "risk". If the result were confirmed, the implications would be considerable. First, it would be possible to shape agricultural production by no longer manipulating price levels but manipulating their variability. In developing countries, for example, it would be possible to obtain major increases in the domestic production of food products without increasing prices, but simply by stabilizing them. A further consequence of this study is that the response of agricultural supply to price changes is probably more complex than is generally believed. In such a context, the dynamic consequences of the farmer's reaction to price changes become very difficult to predict. Moreover, the authors have shown that in a free market a situation of this sort can lead to chaos, with enormous variations in supply, demand and prices. Lastly, if one accepts the results provided by the theoretical model, described in the first part, stabilizing the prices of farm products, even if desirable, is also self-destructive, as nothing can prevent production from increasing indefinitely with stabilized prices. This means we have to look for other ways of regulating agricultural supply. In effect, there still remains plenty of work for the agricultural economists! Jean-Marc Boussard INRA, Paris Françoise Gérard CIRAD, Paris 9 9