Impacts of the Northeast Interstate Dairy Compact on New England Milk Supply

Econometric Analysis of New York Milk in Order 1 Pool, Class I Utilization in Order 1, and Diversions and Transfers to Plants Outside New England[1]

Charles F. Nicholson and Rick Wackernagel

Department of Community Development and Applied Economics

The University of Vermont

Background

The quantity of milk pooled under Order 1 provided by New York producers has increased since the start of the Compact in July 1997. The percentage of milk receipts from producers used in Class I varies seasonally, but average Class I utilization during July 1997 to June 1998 was lower than in all comparable periods during the previous 10 years. Diversions and transfers per month during the compact period have averaged more than twice the quantity of diversions and transfers during comparable annual periods in 1995-96 and 1996-97. These changes provide evidence that the nature of milk flows into the Compact-regulated area (CRA) has changed since the start of the Compact, and that the magnitude of changes in milk flows may be affecting the blend price paid to dairy producers in the CRA.

Graphical analysis of trends is helpful to discern changes in milkshed composition. However, econometric analyses provide complementary information about what factors underlie changes in these three variables, and whether the current patterns are consistent with historical relationships and patterns. Because econometric analyses simultaneously account for many of the factors underlying changes in milk allocation, differences between outcomes predicted by the model and observed values can be attributed to changes in variables not included in the model. Example of such ‘excluded’ variables would be changes in incentives for pooling of milk under the Compact, and the addition of new plants to the Order 1 pool. Thus, the objective of this study is to examine econometrically the allocation of milk in the New England milkshed, and to provide evidence about how the Compact may have affected milk allocation patterns.

Methods

Milk production, allocation to a market, and class use vary with economic and biological factors such as prices, weather, and seasonal changes in consumption patterns. We developed econometric models using monthly data from January 1991 to June 1997 to determine the underlying relationships between prices, weather, and seasonal changes in consumption and the first two variables of interest. A similar model was constructed for examination of diversions and transfers using monthly data from January 1995 to June 1997. Based on the relationships in the model, predictions for the period July 1997 to June 1998 can be made. The predictions use actual values of variables during this period except for the milk-feed price ratio, which uses the Order 1 blend price without the over-order premium. These predictions are compared to the actual values observed during the period since the Compact entered into force. When the actual and predicted values differ a great deal, this is an indication that the current patterns are inconsistent with historical relationships influencing milk flows and allocation in Order 1.

The criteria used to select a specific econometric model among possible alternatives are subjective (Judge et al., 1980). Typically, model development includes decisions about which variables to include and which to leave out, as well as choices about whether models will be linear, non-linear, or linear in logarithms. Economic theory, practical knowledge, and comparisons of results from alternative models guide the choice of variables included. The two most important criteria used to evaluate the models developed for the analyses above are:

a) Explanatory power of the model, that is, how much of the variance in the cow numbers or milk per cow was explained by the included variables (as measured by the adjusted R²);

b) Signs (negative or positive effect) and statistical significance of key variables such as the milk feed price ratio.

The price variable included in the final models is the milk-feed price ratio for New England (an indicator of dairy profitability) lagged by three months to account for biological lags in response to price changes. Rainfall and temperature deviation[2] variables were included to represent the influence of these variables on milk production in New England. Eleven seasonal dummy variables (that is, variables whose value is either 0 or 1) were included to represent seasonal changes in milk consumption and production. A trend variable represented other effects such as increasing milk per cow. The model explaining milk from New York pooled under Order 1 includes a dummy variable for July 1994 to March 1995 to account for a fluid plant in Massachusetts pooled under Order 2 during that time. The model for diversions and transfers includes a dummy variable to account for reduced processing activity at the Hinesburg, Vermont cheese plant as of June 1997.

The model for milk from New York pooled under Order 1 used the Generalized Least Squares (GLS) method (Judge et al., 1980) due to the presence of autocorrelation of the error term[3]. The other models used the Ordinary Least Squares (OLS) method because diagnostic tests (Godfrey, 1978) indicated this was appropriate. The models developed explain much of the variation in the three variables of interest; the explanatory variables account for about 78% of the variance in milk from New York pooled under Order 1, 80% of the variance for the percentage of milk used in Class I products, and 93% of the variance in diversions and transfers (Table 1). Predicted values during the pre-compact period track the actual values closely (Figures 1, 2, and 3). The model predictions allow designation of ‘upper and lower bounds’, or confidence intervals for predicted values, which are necessary because there is a degree of uncertainty in parameters of the econometric model. All actual values during the pre-Compact period lie within the model-predicted confidence intervals. The coefficients produced by the model generally are consistent with prior expectations.

Results

The econometric analyses provide evidence that some of the observed recent changes are inconsistent with relationships influencing milk flows and allocation during 1991 to mid-1997 (Figure 1). The increase in milk from New York producers pooled under Order 1 differs from the pattern predicted by the model. Actual values outside the confidence intervals indicate that the predicted and actual values are statistically significantly different. From November 1997 to June 1998, the quantity of milk received from New York producers is statistically significantly higher than the amount predicted based on historical relationships. The difference between the predicted and actual milk pooled under Order 1 from New York producers totaled 179 million pounds during the eight months from November 1997 to June 1998. The difference between the actual and the upper bound of milk predicted from New York producers pooled under Order 1 totals 50 million pounds during the same period.

Actual Class I utilization of producer milk during November 1997 to May 1998 is lower than that predicted by the model on the basis of historical relationships among variables affecting milk supply and demand in the CRA (Figure 2). However, all actual values of Class I utilization lie within the upper and lower confidence intervals based on the econometric model. Thus, none of the actual values are statistically significantly different from the values predicted by the model. Class I utilization lower than in previous years is in part attributable to higher milk production in New England as a result of higher blend prices under the Compact. In addition, the increase in milk from New York producers pooled under Order 1 probably contributed to lower Class I utilization. Thus, there is some evidence that the blend price in Order 1 may have been somewhat lower than would have been predicted based on patterns of milk allocation during 1991 to mid-1997. The estimated impact of lower Class I utilization on the blend price is small, however, about $0.05 per hundredweight (Nicholson et al., 1998).

The actual amount of diversions and transfers is generally below the amount predicted during the first few months of the Compact period (Figure 3). After October 1997, however, actual diversions and transfers exceeded the predicted amount in each of the months through June 1998. In the months January through June 1998 the actual values are statistically significantly higher than the amount of diversions and transfers predicted based on historical supply and demand relationships. Actual diversions and transfers were 92 million pounds more than the total predicted for the Compact period. The predicted levels of diversions and transfers represent between 5.8 and 9.3% of the total Order 1 pool in the months starting July 1997 (Table 2). The ability of the econometric model to predict historical patterns is quite good, but the model’s predictions during the compact period are much less accurate. This suggests that factors other than basic supply and demand variables are influencing the level of diversions and transfers to plants outside of the New England area.

Conclusions

The above results provide evidence that observed values of milk from New York producers are somewhat inconsistent with what would be expected based on historical relationships among prices, supplies, and demand. In contrast, the observed percentages of producer milk assigned to Class I are lower than predicted by the model for much of the Compact period, but are not so low as to be inconsistent with what would be expected based on historical relationships. The amount of diversions and transfers is higher than the amount predicted in most months, although the difference is statistically significant for the months since January 1998. In part based on the foregoing information, the Compact Commission amended the over-order price regulation in November 1998 to limit payment of the Compact over-order producer price to milk disposed of within the Compact regulated area, but with a seasonally adjusted allowance for diverted or transferred milk. Some of the differences between actual and predicted values are undoubtedly due to changes in incentives under the Compact, but other factors may be important as well. This analysis does not provide evidence about which effects are due specifically to the Compact and which may be due to other factors. The result of the analysis, therefore, should be interpreted with caution.

References

Foley, R. C., D. L. Bath, F. N. Dickinson, and H. A. Tucker. 1972. Dairy Cattle: Principles, Practices, Problems, and Profits. Philadelphia: Lea and Febiger.

Godfrey, L. G. 1978. Testing Against General Autoregressive and Moving Average Error Models When the Regressors Include Lagged Dependent Variables. Econometrica (46):1293-1302.

Judge, G. G., W. E. Griffiths, R. C. Hill, and T-C. Lee. 1980. The Theory and Practice of Econometrics. New York: John Wiley and Sons.

Nicholson, C. F., B. Resosudarmo, and F. W. Wackernagel. Impacts of the Northeast Interstate Dairy Compact on New England Milk Supply. Draft paper for submission to the Northeast Dairy Compact Commission, December 1998. [mimeo]

Table 1. Results of Econometric Models of Milk from New York Pooled Under Order 1, Class I Utilization in Order 1, and Diversions and Transfers

	Dependent Variables
Independent variables	NY Milk in Order 1 Pool	% Class I Utilization	Diversions and Transfers
	(GLS)	(OLS)	(OLS)
February dummy¹	-7.2	-0.4	-2,768.6
	(-2.2)	(-0.5)	(-1.4)
March dummy¹	1.8	0.1	-2,811.0
	(0.4)	(0.1)	(-1.0)
April dummy¹	-6.8	-1.0	421.9
	(-1.2)	(-0.7)	(0.1)
May dummy¹	-2.2	-2.1	1,019.7
	(-0.4)	(-1.5)	(0.2)
June dummy¹	-6.7	-4.2	1,273.0
	(-1.4)	(-3.6)	(0.4)
July dummy¹	-7.3	-3.0	7,079.9
	(-1.8)	(-2.9)	(2.0)
August dummy¹	-8.1	-0.8	10,527.0
	(-1.8)	(-0.7)	(2.8)
September dummy¹	-13.8	3.1	4,272.8
	(-2.6)	(2.3)	(0.9)
October dummy¹	-11.9	3.8	4,460.8
	(-2.1)	(2.7)	(1.0)
November dummy¹	-14.4	4.5	2,070.7
	(-2.9)	(3.5)	(0.6)
December dummy¹	-5.0	1.2	-503.2
	(-1.3)	(1.2)	(-0.2)
Lagged Milk feed price ratio (3 month lag)²	-0.02 (-0.005)	-3.1 (-3.3)	-15,768.1 (-3.7)
Square of previous summer rainfall	-	-	31.6
(inches)	-		(2.3)
MA fluid plant dummy ³	-22.0	-	-
	(-10.4)	-	-
VT cheese plant dummy ⁴	-	-	24,269.0
	-		(7.0)
Trend	-0.2	-0.05	-286.0
	(-5.0)	(-5.9)	(-2.5)
Constant	138.7	55.1	51,164.1
	(17.6)	(27.4)	(4.1)






Model evaluation characteristics
Adjusted R²	0.73	0.76	0.83
LM test statistic first-order autocorrelation, OLS model⁵	30.64	1.81	0.002
Significance of first-order autocorrelation⁶	0.01	0.99	0.99
Number of observations	75	75	30
Data period	January 1991 – June 1997	January 1991 – June 1997	January 1995 – June 1997

Note: Figures in parentheses are t-statistics.

¹ Dummy variables equal one or zero. For observations in February, the February dummy equals one. For all other months the February dummy equals zero. The dummy variables for the other months are defined analogously.

² Equals the ratio of Order 1 Blend price per hundredweight divided by the Vermont grain price three months prior to the current month. For example, the value of this variable for April would equal the milk-feed price ratio for the previous January.

³Dummy variable for period in which the Agawam, Massachusetts milk plant was pooled under Order 2. Values of the variable are 1 when the plant was pooled under Order 2 (July 1994 to March 1995) and 0 when the plant was pooled under Order 1 (January 1991 to June 1994 and April 1994 to June 1998)

⁴Dummy variable for the operation of a cheese plant in Hinesburg, Vermont. Values of the variable are 1 when the plant was not operating (June 1997 to June 1998) and 0 when the plant was operating (January 1995 to May 1997).

⁵ Lagrange multiplier (LM) test for first-order autoregressive or moving average processes (Godfrey, 1978). This statistic is distributed as a c² with the number of regressors minus one degrees of freedom.

⁶ Probability of first-order autocorrelation based on LM test from the OLS model.

Table 2. Actual and Predicted Diversions and Transfers of Fluid Milk Products to Plants Located Outside of New England, July 1997 to June 1998

	Diversions and transfers			% of total Order 1 Pool
Month	Actual (mil lbs)	Predicted¹(mil lbs)	Total Order 1 Pool (mil lbs)	Actual	Predicted²

July 1997	34.4	39.4	462.9	7.4	8.5
August 1997	29.5	42.9	463.2	6.4	9.3
September 1997	33.0	37.0	441.2	7.5	8.4
October 1997	31.1	40.3	452.7	6.9	8.9
November 1997	43.6	37.1	450.8	9.7	8.2
December 1997	40.3	34.5	481.9	8.4	7.2
January 1998	42.6	32.8	493.9	8.6	6.6
February 1998	49.8	28.3	453.1	11.0	6.3
March 1998	50.8	27.6	505.7	10.0	5.5
April 1998	48.5	30.8	495.3	9.8	6.2
May 1998	53.3	30.4	521.7	10.2	5.8
June 1998	61.5	30.0	497.3	12.4	6.0
Total during Compact period	518.4	410.9	5,720.0	9.1	7.2

¹ Prediction is from the econometric model estimated using monthly data from January 1995 to June 1997.

² Equals predicted value of diversions and transfers divided by total Order 1 pool times 100.

[1] A previous version of this document was submitted to the Compact Commission in September 1998 as testimony regarding the proposed rule on diversions and transfers of milk.

[2] The variable for temperature deviation is defined as the square of the difference between the actual mean monthly temperature and 50 ° F, which is the mean of the temparature range considered biologically optimal for milk production (Foley et al., 1972).

[3] Autocorrelation exists in an econometric model when the error term in one period is related to the error term in other periods. Although the coefficients estimated with OLS are unbiased in the presence of autocorrelation, the estimated variance of the coefficients is biased. The GLS method adjusts for this bias, and is appropriate when statistical tests such as Godfrey’s (1978) indicate that autocorrelation is a problem.