Chapter 2 narrowed the BFP options available for further analysis from 32 down to 11. The purpose of this chapter is to subject each of these 11 options to as much econometric analysis as is possible within the constraints of the data available for such analysis. Recognizing that there may be an interest in seeing how the options excluded in step 1 performed under the same analytical rigor, Appendix D provides the same detailed analytical results for all of the 32 options.

Chapter 1 set forth the three criteria utilized in the step 2 analysis. In summary form, these included:

• Reflect national supply-demand conditions for manufactured products. The extent to which the option reflects national supply-demand conditions for manufactured products was designed to isolate the effects of manufactured product stocks as a measure of supply-demand balance on the prices generated by each option.

• Reflect changes in the value of milk used in manufacturing. The extent to which the option reflects the value of milk for manufacturing was designed to measure how well processor receipts from manufactured products are reflected in the milk price for each BFP option analyzed. This was done sequentially to assess the combined impact of changes in cheese, butter and NDM prices as well as separately to obtain the effect of each product on the price of milk.

• Provide price stability. The extent to which the option generates relative price stability. This criterion arguably becomes a more important consideration in the absence of a price support program. Therefore, how much price variability each option generates was measured both in terms of price movements over time and the price changes resulting from shocks in stocks.

Many different ways of evaluating the 11 BFP options were discussed and evaluated. The desire of the USC was to select an analytical procedure that could be applied uniformly across as many of the options as possible within the time constraints imposed by the decision process set forth by the Secretary. In addition, USC wanted a technique that would require few assumptions on how participants would perform under alternative BFP options. The procedure selected is referred to as vector autoregression (VAR). This procedure is designed to analyze the relationship of economic data over time, otherwise referred to as time series analysis. Although VAR is not the only time series methodology, it is a very flexible analytical tool.

VAR is particularly useful in this study because it allows consideration of feedback effects between milk prices and product prices. That is, it is well known that product prices affect milk prices. But logically, the price processors pay for milk also affects product prices. These interrelationships are considered utilizing the VAR technique. Moreover, utilizing VAR allowed us to analyze the reality that product prices not only affect milk prices this month but the next month and, perhaps, the following month. In addition, we use VAR to simulate the impacts of a change in important variables, such as manufactured product stocks, on prices over time. The basis for these predictions is the historical regularity between prices and stocks.

VAR can be utilized either in terms of levels or changes. For example, VAR analysis of levels could relate the price of milk to the price of products. Alternatively, VAR could be utilized to analyze just the impacts of changes in the price of milk on changes in the price of products which is referred to as first difference analysis. USC utilized the first difference approach because it more directly focuses, for example, on how the price of milk might be expected to change if the price of cheese changes or if cheese stocks change. (As a check, USC did all analyses based on price levels. We found no difference in the results. Both analyses are included in Appendix D. In addition, we checked standard tests for nonstationary (Dickey-Fuller tests and augmented Dickey-Fuller tests) on levels and found each basic formula price candidate to be nonstationary in levels and stationary in first differences. These results favor using the first difference approach.) It is important to note that all stocks changes are measured in terms of total solids equivalent where the milk equivalent of milkfat is weighted by 40 percent and the milk equivalent of solids-not-fat is weighted by 60 percent.

Obviously, there is interest in the price level as well as in the impacts of price changes. As a point of reference, therefore, Table 11 presents the mean prices that were generated by each of the 10 of the 11 BFP options as compared to the M-W and the adjusted M-W series. It will be noted that the mean price levels range from $11.35 per cwt. to $12.45. These mean prices have relevance since, for example, if either A/B price series were implemented to set the level of price over time, the result would be increased production of milk, which over time would force the price back down. Alternatively, the BFP could be used as a price mover. In other words, the price of milk would change from the level at which it is implemented by the amount of change in the BFP. Our first difference analysis is of this latter type. No attempt was made to determine the mean price under the no BFP option. As indicated in Chapter 2, the price for milk used in manufacturing would be market determined and would vary regionally.

Limitations of VAR

The VAR technique is not without controversy. Economists often prefer to set up a structural model based on economic notions of how the world ought to operate. Such models generally assume that firms maximize profits and consumers maximize satisfaction. VAR makes no such assumptions. Instead, it takes the data for what it is and analyzes the relationships over time, based upon past experience.

While USC considers the VAR approach to be appropriate for this study, it can only provide a rough guide to how the BFP options would be expected to perform in the future. Simulating over a short past time period (1991-95) is not the same as asking how would the world be different if another BFP option were in place. That is, the path of prices and stocks would have been affected by the change in policy. While less likely, the stocks/price relationships could also change.

Therefore, it is important not to put too much emphasis on any individual statistic or result. Further, some of the statistical differences probably are not economically significant. However, when one BFP option performs consistently better than another, greater reliance can be placed on the results, particularly when supported by common sense reasoning.

Reflection of National Supply-Demand Conditions

As implied by the preceding discussion, the methodological considerations in measuring how well the BFP options respond to national supply-demand conditions for milk used for manufacturing are complex. USC utilized changes in the combined total of public and private sectors of manufactured product stocks (measured in total solids equivalent) as the indicator of supply and demand conditions. Stocks of butter, NDM and cheese reflect the residual of the interaction of both supply and demand forces as processors make decisions on what share of their production they can profitably sell or store for future sale. In contrast with the past, government storage has declined in importance over the period 1991-95 to the point where they are currently nonexistent (Figures 19-21). This has happened because reductions in the milk price support level have made the Commodity Credit Corporation (CCC) a relatively unattractive market compared with private sector sales or storage. We isolated the critical turning points for this change in policy as being May 1992 and November 1994. These critical turning points were accounted for in the VAR analysis by isolating the effects of stocks on the price of milk before and after these turning points.

Four statistical procedures were used to determine the extent to which the BFP options reflect supply-demand conditions as compared to the M-W and adjusted M-W series (Table 12). Each procedure measures the effect of a 248 million pound total solids milk equivalent increase or decrease in stocks on price (this number reflects on average one standard deviation shock in stocks over the study period). These measures could be applied to all of the options except pooling differentials with no BFP.

• Was there an inverse relation between stocks and the BFP? That is, when stocks were increased, did the milk price decline? The answer was yes for all of the options except the Grade A/B and adjusted Grade A/B options. That is, both A/B series prices increased with an increase in stocks. This result is interesting since both the M-W and adjusted M-W price declined with an increase in stocks. We have two possible explanations for this perverse performance: (1) The A/B prices are never directly experienced in the marketplace. Therefore, milk industry participants never make decisions based upon either series (at least not directly). (2) Private stocks are estimated, not audited. The estimates may not be completely accurate, thus distorting the outcome when the explanatory power is quite small.

• What percentage of the variation in the monthly price is explained by stocks? The explanatory power of stocks ranged from 25 percent for pricing components with no BFP to less than 1 percent. For the butter/powder-cheese formula with seasonal yields weighted by U.S. milk production and the price support make allowance, stocks explained 14 percent of the price variability. The addition of the feed cost snubber to the product price formula materially reduces its performance. The results clearly indicate that stocks appeared to explain a larger proportion of the price variation in product price formulas when weighted by U.S. production.

• What is the cumulative influence of the 248 million pound change in stocks on the change in price at the end of 12 months? Here, pricing components with no BFP and the butter/powder-cheese formula with the cost-based make allowance and seasonal yields had $0.25 per cwt impact. For all of the other options a 248 million pound increase or decrease in stocks had less than $0.10 per cwt price impact. The A/B series had the wrong direction of impact. That is, as stocks increased, price increased.

• What percentage of the monthly BFP price variation is explained by variation in stocks at a 12 month horizon? Three of the options stand out in this case as having large proportions of their variability explained by changes in stocks. The butter/ powder-cheese formula with seasonal yields and a cost-based make allowance weighted by U.S. production had 10 percent of its variation explained by changes in stocks and pricing components with no BFP had 7.6 percent explained. The remaining options explained less than 3 percent of the variation.

Three important conclusions arise from this analysis:

• Pricing components with no BFP performed consistently better than any of the other options. It is believed that this better performance is a result of the way the component values are derived under this BFP option. The protein prices are derived from the weighted monthly average 40# block cheese price, butter from the weighted monthly average Grade AA butter price and the value of other solids as a residual of the nonfat milk (NFD X yield minus the protein value divided by 5.4 pounds of other solids). Since cheese prices, butter prices, and nonfat dry milk prices each respond to stock levels (as cheese stocks start building, cheese is offered on the National Cheese Exchange, and cheese prices decline, and so does the protein value), component values likewise respond immediately to changes in product prices. Therefore, the full impact of these product price changes are reflected in component values and are not impacted by annual yields weighted by MN and WI or by U.S. milk production, as is the case for the product price formulas.

• The two butter/powder-cheese formulas with seasonal yields performed equally well.

• The A/B pricing options do not respond to an increase in stocks as expected based on economic logic. That is, as stocks increase prices increase, which is contrary to our expectation. This is viewed by USC as a substantial strike against the A/B options.

• The addition of a feed cost snubber on a product price formula substantially reduces the extent to which the formula reflects national supply and demand conditions.

When product prices change, the BFP should adjust reflecting both the magnitude of change in product prices and the share of the product's sales in the mix of manufactured products. VAR was utilized to measure the proportion of variation in the BFP for each option that is explained by the prices of cheese, butter and NDM. Sequentially, it was found that cheese prices have the largest price impact for all options, followed by butter and then NDM. Table 13 provides two related methods for measuring the extent to which the BFP options reflect values of milk for manufacturing:

• The proportion of BFP variability in price changes explained by all three products. The combination of cheese, butter and NDM explained from 21 percent to 58 percent of the BFP price variability. Pricing components with no BFP performed best in reflecting product values by having 58 percent of its variations explained by product prices. The butter/powder-cheese options with the California cost-based make allowance had 41-45 percent of variation explained by product prices -- substantially higher than most of the remainder of the options which generally explained about 25 percent of the variability. As might be anticipated, for the cost-snubbed product formula product prices explained only 5 percent of the BFP price variability.

• Proportion of price variation explained by individual products. The results for the sequential explanatory power of product prices are robust in the sense that with one exception, the product formulas with the California cost-based make allowance and the pricing component with no BFP options have a consistently strong relationship between the product prices and the computed milk price. The exception is the relationship between the pricing components option and the price of cheese, where the variation explained in the milk price is only 23 percent. In this case, butter prices explain more of the variability than cheese. The reason is that the price of butter has been more volatile than cheese prices (ranging from $0.70 to $1.12 per pound). In the competitive pay price options and product price formula options cheese, butter and nonfat dry milk yields are weighted by either MN and WI production or by national production. Cheese by far gets the greatest weight. But with the component price option, there are no weights, simply a price per pound of component derived from one pound of product (cheese, butter and nonfat dry milk).

The conclusion the USC draws is that product price formulas with a cost-based make allowance and pricing components with no basic formula price generally do the best job of reflecting product values in the price of milk. Competitive pay prices consistently rank in the lower half of the options. The feed cost snubbed product price formula performed the poorest in reflecting product values. This should not be surprising since the addition of feed costs logically would be expected to reduce the relationship of product prices to milk prices.

Stability of BFP Options

With milk prices being more unstable in the 1990s, and in the absence of a support price after 1999, greater attention might logically be given to the amount of price variation experienced by each option. Two statistical measures were utilized to measure stability (Table 14):

• The standard deviation of the milk price over the period 1991-95 measures the amount of unexplained price variability about the mean. The range in standard deviation was from $0.66 to $0.95/per cwt. The least variability was experienced by butter/powder-cheese formulas with annual yields weighted by U.S. production and utilizing the California cost-based make allowances. Pricing components with no BFP experienced the second largest level of price variability. Butter/powder-cheese formulas with seasonal yield adjustment and the price support make allowance generally had higher levels of price variability.

• Price variability at 12 months. Based upon the VAR results between the first differences of each candidate price and the first differences of milk stocks, we can measure the uncertainty in price differences at successive months into the future. One would expect a stable price formula to show lower variability in first differences as we look 12 months into the future. The range in standard deviation was from $0.45 per cwt to $0.69. Pricing components with no BFP experienced a standard deviation of about $0.45 per cwt, but several options were in the $0.45 to $0.50 per cwt range.

USC draws the conclusion that some of the BFP options indeed are more stable than others. More sophisticated options having cost-based make allowances appear to have a particularly favorable price stabilizing effect. Perhaps this is because the option itself incorporates key values within the price that would otherwise have to be revealed in the marketplace as potentially destabilizing factors.

Table 15 summarizes the results of the Step 2 analysis by ranking each BFP option by each of the three criteria and their subcomponents. Table 16 boils these rankings down by weighting each of the criteria equally. While small differences in these ranks may not be meaningful from an economic perspective, this procedure and weighting suggests substantially superior performance for pricing components with no BFP, and for the butter/powder-cheese formulas that utilize a cost based make allowance. While not based on empirical analysis, USC would expect the pooling differentials with no BFP to perform more like the competitive pay price options than either the product price formulas or pricing components with no BFP.