5 Using the economic surplus model to measure potential returns to the research investment

The model

An economic surplus model (Alston et al 1995) was used to measure the potential returns to the research on genetic improvement of dual-purpose millet and sorghum. A partial-equilibrium, comparative static model of a closed economy was used in the analysis. With relatively bulky, perishable commodities such as meat and milk, a closed economy model which assumes relatively little international trade is appropriate.

The adoption of a cost-reducing or yield-enhancing technology increases the supply of a commodity such as meat or milk. Because there is little or no international trade, the increase in supply reduces both the price of the commodity to consumers and the cost to producers.  The simple case of linear supply and demand curves with parallel shifts was chosen. A review of studies of research benefits by Alston et al (1995) reveals that the majority of such studies use similar assumptions. Alston and Wohlgenant (1990) argue that when a parallel shift is used, as suggested by Rose (1980), the functional form is largely irrelevant, and that a linear model provides a good approximation to the true (unknown) functional form of supply and demand.

In Figure 5.1, D is the demand function for the product (meat or milk) and S₀ is the supply function for the product before the research-induced technical change (e.g. an improved variety). The initial equilibrium price and quantity are P₀ and Q₀, respectively.  Adoption of the new technology shifts the supply curve of meat or milk to S₁, resulting in a new equilibrium price and quantity of P₁ and Q₁, respectively. Gross annual research benefits are measured by the area between the two supply curves and beneath the demand curve.

Figure 5.1. Measuring gross annual research benefits (change in total surplus).

This area represents the total increase in economic welfare (change in total surplus), and comprises both the changes in producer and consumer surplus resulting from the shift in supply. Consumers are better off because they consume more at a lower price. Although producers are receiving a lower price for their milk or meat, they are able to sell more, so their benefits increase, unless supply is perfectly elastic or demand is perfectly inelastic, in which case their revenues remain the same (Figure 5.1). The algebraic derivations of these surpluses are shown in Table 5.1. The change in total surplus can be thought of as the maximum potential benefits to a technology (an improved crop variety); they would be actual benefits if the research were successful and its output fully adopted.

Outputs from the feed simulation model—the percentage increase in milk and meat output resulting from improved quality straw—provided the estimate of the proportionate increase in productivity per head, E(Y). E(Y) was thus measured empirically in terms of quantity (i.e. the horizontal shift in the supply curve, or distance eb in Figure 5.1), and translated into a common currency (US dollars) by calculating the distance ac in Figure 5.1 (the vertical shift in the supply curve). Productivity gains were then converted to gross proportional reductions in cost per tonne of output (C) by dividing the estimated productivity gain by the elasticity of supply. This is a gross reduction in output cost, because the change in input costs (E(c)) associated with the introduction of the new varieties also have to be considered. These include any additional variable input costs (e.g. labour,  fertiliser) that are necessary to achieve the higher potential yields of the new technology. The net proportionate change in marginal cost per tonne of output (k) is derived by subtracting the effect of variable input cost changes associated with the use of the technology.

Assessing the returns to genetic improvement in dual-purpose millet and sorghum research

Gross annual research benefits, measured by the change in total surplus, represent the maximum possible potential benefits from a new technology. To estimate the potential net benefits accruing to current research, however, some uncertainties must be considered: the uncertainty surrounding if and when the research may be successful, the uncertainty in the proportion of farmers adopting and the rate at which they may adopt the new varieties.   The economic surplus model accounts for such uncertainties by using probabilities.  One of the challenges in using the economic surplus model to measure the potential returns to research, therefore, was how to estimate research and adoption lags, probability of research success, and the ceiling level of adoption. The approach taken to estimate these uncertain parameters is described on page 27.

Figure 5.1 represents research benefits for one year. A successful research investment will yield benefits over a number of years. As the level of adoption increases there will be a further shift in the supply curve, and a corresponding change in benefits. This adoption process was assumed to follow a typical S-shaped curve approximated by a discrete time distribution (Jacobsen and Norton 1996) (see Figure 5.2).

Figure 5.2. Research adoption time frame.

Since the results of research are likely to depreciate over time (e.g. due to availability of newer technologies), a depreciation factor also needs to be taken into account in the calculation of net benefits. It is assumed that an improved variety will be a relatively sustainable technology (compared to a pesticide against which the pests can eventually develop resistance, for example). Thus it was assumed that the benefits would not depreciate substantially (1% per year), and that this depreciation would not commence until 10 years after the new varieties became available.

The benefits and costs of the research were arrayed on a yearly basis over a 30-year period, and a discount rate of 5% applied to calculate the net present value (NPV) of the research, as the sum of total discounted returns minus total discounted costs. A positive NPV implies a research programme that is profitable. The internal rate of return (IRR), or the discount rate at which the NPV is zero, was also calculated. Using this criterion, research programmes are profitable if the IRR is greater than the opportunity cost of funds.  The benefit:cost ratio, or total discounted returns divided by total discounted costs, was also calculated. Since many of the baseline assumptions are debatable, sensitivity analyses were undertaken to assess the effect of different discount rates, adoption levels, research timing and cost assumptions, and probability of research success on the NPV, IRR, and benefit:cost ratio.

Data and assumptions used in the economic model

Quantities and costs associated with the new varieties

The pre-research quantities (Q₀) of meat and milk produced in RD1 (north) and RD2 (south), identified in the GIS analysis (see Chapter 3), were calculated and used in the economic analysis. Annual milk production in RD1 is 7.96 million tonnes. Each year, 8.26 million tonnes of milk are produced in RD2. Beef production is estimated at 104,000 t in RD1 and 173,000 t in RD2 (Table 3.1). The percentage increase in meat and milk production made possible with improved quality straw (E(Y)), estimated from the feed simulation model, was applied to these initial quantities of meat and milk being produced in each domain. An important assumption is that additional input costs associated with the new varieties (E(c)) will be negligible (so E(Y) = k). This is because the increase in quality and quantity of fodder we are evaluating is assumed to be due to genetic improvement rather than improvements in management (e.g. increased fertiliser or other inputs). This assumption will have implications for the estimated adoption rate (see page 27).

Prices and price responsiveness (elasticities)

Average 1997 farm-level prices of livestock outputs were estimated from the survey data.  Elasticities of supply and demand were taken from regional empirical studies (see Table 5.2).

Research lag, costs and probability of research success

A survey of sorghum and millet breeders was undertaken to elicit estimates of the likely length of time needed to develop and release new dual-purpose varieties, and the probability of being successful within that time frame (Table 5.3). Fifteen breeders from ICRISAT and five Indian research centres were interviewed. The economic model used scientists’ most likely estimates of the research lag, costs and probability of research success for the baseline analysis.

Estimates of the research lag, i.e. the number of years until a new variety is released from ICRISAT ranged from 4 to 12 years, with an average ‘most likely’ research time frame estimate of 7.6 years. The first 2 years of research aimed at identifying traits and markers were assumed to take place at ICRISAT. Four NARS collaborating research centres were assumed to begin complementary research activities in year 3 for a period of 8 years.  Release of new varieties was thus predicted to happen in year 8, followed by 2 years of testing before release to farmers by private sector seed companies and/or the public sector.

This estimated time horizon is shown in Figure 5.2.

The research costs corresponding to these assumptions were estimated at US$ 0.66 million in the first year (this includes up-front equipment costs); these costs are lower in the second year, and rise in the third year when the NARS start up their activities and thus have some initial fixed costs (Table 5.4 and Appendix B). For years 7–10, research costs only include the NARS costs (see Appendix B).  With the new varieties becoming available to farmers in year 10, the costs included for years 11–16 are meant to reflect the costs of dissemination of this new technology via the extension service. While this is a difficult figure to estimate, Evenson and McKinsey (1991) argue that in most developing countries, the amount spent by governments on extension is typically three times the amount spent on research. Thus, we tripled the estimate of the NARS research costs to reflect the investment being made in extension in a country with so many smallholder farmers to reach.

Estimates of the probability of the research being successful, i.e. new varieties being released in this time frame ranged widely from 20% to 90%, with an average ‘most likely’ estimate of 54% (Table 5.3). In the economic analysis, this probability of research success is multiplied by the ‘maximum’ potential benefits to arrive at an ‘uncertainty-adjusted’ benefit estimate (so by using a probability of research of 50% in our baseline economic analysis, we are essentially halving the potential benefits). This probability of research success will likely change over time as new information becomes available—for example, if the same breeders were interviewed a year later, they may be more or less optimistic of achieving success, given what they have learned about the research problem in the meantime.  

Adoption lag and ceiling level of adoption

Two approaches were taken to estimate probable adoption lags and maximum adoption levels for improved dual-purpose sorghum and millet within the recommendation domain.  First, the same breeders interviewed above were asked, given their past experience with dissemination of comparable improved varieties, to give their estimates of the likely length of time until maximum adoption is achieved, and the maximum adoption rate at the end of that period (i.e. percentage of farmers adopting) (Table 5.3). The researchers estimated that the adoption lag would range from 4 to 8 years, with an average of 5.8 years. The predicted ceiling levels of adoption ranged from 30% to 70% of farmers, with a mean of 46%.

Since breeders may understandably tend to overestimate potential adoption rates, the second approach taken was to review recent adoption literature. Bantilan and Joshi (1996) studied the adoption of the disease-resistant  pigeon pea variety ICP 8863. They found that

maximum adoption levels reached 51.8% of total pigeon pea area in Rangareddy District, Andhra Pradesh, and 58.7% in Osmanabad District, Maharashtra. Maximum adoption levels reached 58.9% in Karnataka and 17.2% in Maharashtra. In this particular study, the adoption lag was 7 years.

In another case study of adoption of new higher grain-yielding pearl millet varieties and hybrids (Bantilan and Joshi 1994), ICRISAT social scientists noted that the combined adoption rates of a new class of varieties was as high as 75%. In this case also, the adoption lag was 7 years. But these may be cases of highly successful technologies, since many other high yielding varieties (HYVs) released have not been so successful achieving widespread adoption. Kiresur (1992) studied the adoption behaviour of new grain sorghum varieties in Karnataka by district. He noted high adoption rates in districts with relatively little sorghum. However, in the traditional sorghum districts, maximum adoption levels ranged between 3% and 48%, with a state average of 23%, and an adoption lag of 23 years.

In our analysis we are referring to adoption by farmers within our RD, i.e. farmers that are likely to be using dual-purpose sorghum and millet varieties and would be interested in sowing improved ones. However, it was beyond the scope of this study to carry out a large-scale farm household survey within our chosen RD to determine constraints to adoption of the new varieties (e.g. lack of markets for livestock outputs, lack of access to the new varieties etc). We, therefore, are assuming an adoption lag of 6 years and a conservative ceiling level of 10% in the baseline analysis. Keeping in mind that dual-purpose varieties are already in widespread use by farmers in the RD, and the new varieties will involve negligible additional monetary costs (higher straw yields could involve more labour inputs), it is possible that the ceiling level of adoption could be much higher than 10%. Thus a sensitivity analysis was carried out, increasing the maximum adoption levels to 30% and 50%, and examining the implications for estimated research returns.

Results of the economic model: Returns to the research investment

Estimation of potential maximum, or gross, annual benefits in terms of increased value of meat and milk production as a result of higher quality millet and sorghum fodder is shown in Table 5.2. This assumes the research product is available and fully adopted. The estimated potential productivity gains could result in a 3% to 6% reduction in the cost per tonne of producing milk (depending on the region), with a 6% to 7% reduction in the cost per tonne of beef produced (Table 5.2). The lower cost of production results in an increase in the amount of beef and milk supplied by farmers and a lower price to consumers.

The change in total surplus (Table 5.2) was adjusted by the levels of adoption, the probability of research success, and a depreciation factor to estimate the returns to genetic improvement of dual-purpose varieties of sorghum and millet (Tables 5.4 and 5.5). These ‘uncertainty adjusted’ benefits, generated over the next 30 years, were then compared to the

research costs and discounted (using a discount rate of 5%) to calculate the NPV of the research.

The net benefit stream (i.e. benefits minus costs over the next 30 years) is shown in Figure 5.3. The NPV of the research is estimated to be US$ 42 million, with an IRR of 28%, and a benefit:cost ratio of 15:1.

Figure 5.3. Predicted net annual benefits from research.

Sensitivity analysis

An ex ante analysis attempts to measure the impact of research that is ongoing, with uncertain timing of outputs and ultimate adoption of those outputs. It is thus critical that an analysis is made of the sensitivity of the final results to some of the assumptions or estimates that are inputs to the economic model (Table 5.6).

Probability of research success

If a research breakthrough were to occur tomorrow—for example, a marker strongly linked with one gene controlling a critical fodder quality trait was identified—we would want to increase the probability of research success from the 50% used in the baseline analysis.  If the probability of research success (keeping the other assumptions constant) was increased to 90%, the NPV of the research increases from US$ 42 million to US$ 78 million, the IRR becomes 35%, and the benefit:cost ratio almost doubles to 27:1.

Adoption levels

Assumptions about the maximum level of adoption of the new varieties strongly influence predicted returns. Since we started with such a conservative ceiling adoption estimate of only 10% of farmers, this was increased to 30% and 50% to examine the implications for predicted research returns. At a ceiling adoption level of 30% of farmers within the RD, predicted returns to research triple to US$ 124 million, with an IRR of 38% and a benefit:cost ratio of 42:1. When adoption reaches 50% of livestock farmers within the RD, research returns quadruple to US$ 208 million, with an IRR of 43% and a benefit:cost ratio of 69:1.

Research period

Breeders are interested in the impact on returns of changes in the estimated length of time the research takes until a new ICRISAT variety is released. This analysis includes in the research lag the 2–3 year period between release and the point at which the variety actually starts reaching farmers. This total ‘research lag’ was conservatively estimated at 10 years in the baseline analysis. When this is increased to 12 years, returns fell from US$ 42 million to US$ 34 million, the IRR decreased from 28% to 24%, and the benefit:cost ratio declined from 15:1 to 12:1. This analysis indicates that predicted returns are much more sensitive to changes in the levels of adoption, for example, than they are to changes in the estimated research period.

Research costs

Research costs were increased by 25% (from a total of US$ 3.95 million over 16 years to US$ 4.94 million), resulting in a slight impact on estimated research returns. The NPV fell by US$ 1 million to US$ 41 million, the IRR decreased to 26%, and the benefit:cost ratio fell to 12:1.

Discount rate

The choice of appropriate discount rate has a significant impact on the results of this model.  The discount rate is a time preference concept. If a ‘socially optimum’ discount rate actually exists, it is evident that such a rate can never be precisely known because the preferences and circumstances of future generations remain unknown (Goodland and Ledec 1987).  Economists disagree as to whether the appropriate social discount rate should reflect the alternative value of public resources being consumed or invested (Alston et al 1995). They do agree, however, that in this type of analysis the rate should be a real rate of interest (adjusted for inflation), and that it should reflect any restrictions placed on alternative uses of the funds. Alston et al (1995) argue that this corresponds to a long-term, risk-free rate of return, such as the real yield from long-term government bonds (typically around 5%, used in the baseline analysis). We assessed the effects of the discount rate on returns by increasing it from 5% to 10% (commonly used in project analysis). The NPV and benefit:cost ratio fell to US$ 16 million and 8:1, respectively.