3.1 Conceptual model

Based on the depiction of decisions from the shade-grown coffee farmers and drawn from Batz et al. (Reference Batz, Albers, Ávalos-Sartorio and Blackman2005), a stylized dynamic model of annual farmer decisions in the presence of future price uncertainty is developed to predict abandonment decisions.Footnote ¹ Using the current time period's observed price and expectations of future prices, and considering the impact of current decisions on future yields, the farmer makes a dynamically-optimal one-period production decision that generates a level of yield and income in that period. The farmer sees the current coffee price and forms expectations of uncertain future coffee prices while recognizing that prices are serially correlated, meaning the current period price depends on prices in the previous period (i.e., farmers observing low prices now expect low prices in the near future). The farmer makes production decisions to maximize the expected net present value of his stream of profits, subject to annual subsistence, wealth (or savings) and labor constraints, and to evolution of yield over time. Although future yields are not uncertain in this framework, the farmer recognizes that decisions in the current period affect the level of yield in future periods because yield declines as a deterministic function of failure to perform coffee bush maintenance. Featuring yield as a function of past actions creates a non-Markov yield evolution – the level of yield in the following year depends on previous maintenance – and generates path-dependence of farmer actions and outputs. With path-dependence in yields, serially correlated uncertain price paths, and irreversible abandonment of coffee farming, the farmer solves a complicated stochastic dynamic optimization problem in each period to determine the dynamically-optimal production decision for the current period, while considering past maintenance efforts, current price and all possible future pathways for prices and decisions.

More specifically, in each year, the farmer decides how to allocate labor between coffee production and off-farm employment: harvest and maintain coffee bushes, harvest only, or abandon production. Maintaining coffee bushes helps sustain yields but is time consuming. Farmers who harvest and maintain (HM) spend their entire labor allotment completing these activities and do not work off-farm. Those who harvest only (HO) and forgo maintenance in the current time period use their remaining labor hours for working off-farm and earning wages. Each period of no-maintenance depresses future yields and sets the farm's productivity on a different path than occurs with maintenance – the transition to an alternate path of yield productivity demonstrates path dependence of the farmer decisions. If coffee production is no longer profitable, the farmer abandons (A) and devotes his entire labor allotment to off-farm work. Stakeholder interviews revealed that repeated lack of harvest and maintenance causes damage to coffee bushes from which yields cannot recover without substantial investment. Because farmers do not have access to such funds, abandonment becomes an irreversible decision (e.g., Wernau and Whelan, Reference Wernau and Whelan2018; Terazono et al., Reference Terazono, Webber and Schipani2019). Unlike annual agricultural production decisions, the farmers' decisions in any time period must consider the impact of those decisions on future values and opportunities, such as the irreversibility of abandonment and the negative impact on future yield of forgoing maintenance. The farmers' decisions are further altered when considering the impact policies have on costs and prices.

3.2 Farmer decision model

The farmer selects one of the three labor/production decisions in each time period, t, while knowing that period's price, p_t. The farmer's per period profit function is:

\[\pi = v({k,p} )= [{pq(k )+ wl(k )- c(k )} ] ,\]

with ${k_t}$ representing the chosen action from the set ${K_t} = H{M_t}, H{O_t}, {A_t};$ q(k) being the amount of coffee produced with action k; w being off-farm wage; l(k) the amount of labor in off-farm wage for that action choice k; and c(k) the costs of action k. To reflect observations and to simplify the solution method, and with ${\beta ^s}$ as the discount factor to time s, the farmer is forward-looking over a finite number of years, T. In every specific time period, s, the farmer solves the stochastic dynamic optimization problem for the current period s while considering T years, subject to a set of per-period constraints and equations of motion for the state variables (described below); that period's known price, p_s; that period's wealth, W_s; expectations about uncertain and stochastic future prices, p_t; and future wealth, W_t:

(1)\begin{equation}\mathop {\max }\limits_{{k_s}} {V_s} = \left\{ {\mathop \sum \limits_{t = s}^{t = s + T} {\beta^s}\textrm{{E}}[{v({{\pi_{t }}({{p_t},{W_t}} ),{p_t}} )} ]} \right\}|{p_{s,}}{W_s}.\end{equation}

The farmer's chosen action has implications for current and future yield, labor, cost, subsistence, wealth, and irreversibility constraints. The model accounts for the path dependence of current yield as a function of maintenance activities completed in prior years through the yield growth function (equation (2)). Performing maintenance in past years leads to higher current yield (Bustamante et al., Reference Bustamante, Isaza, van Heeran, Torres and Romero2009). If, however, the farmer forgoes maintenance to earn off-farm wages to meet subsistence needs, yield in the subsequent year is reduced. Abandoning in the prior year results in no yield for all subsequent years. The logistic yield function with slower growth from lack of maintenance is:

(2)\begin{equation}q({k_{t}}) = \left\{ \begin{array}{@{}ll} {q(k_{t-1})\left[1 + \gamma \left(1 - {\displaystyle{{q({k_{t - 1}})} \over {\overline q }}}\right)\right]}&{\textrm{if}{k_{t - 1}} = H{M_{t - 1}}}\\ {q(k_{t-1})\left[1 + \gamma m\left(1 - {\displaystyle{{q({k_{t - 1}})} \over {\overline q }}}\right)\right]}&{\textrm{if}{k_{t - 1}} = H{O_{t - 1}}}\\ 0&{\textrm{if}{k_{t - 1}} = {A_{t - 1}}} \end{array} \right.,\end{equation}

in which $\bar{q}$ is the maximum yield or the yield carrying capacity. The intrinsic growth rate, $\gamma$, is reduced by a factor of m if the farmer performed no maintenance in the previous year, with no yield uncertainty. Both the non-Markov property of yield's dependence on prior maintenance and abandonment's irreversibility introduce path-dependence into the farmer's decisions. Assuming a binding per-period labor time constraint,

(3)\begin{equation}{L_{t}} = l({{k_t}} )\quad \forall t,\end{equation}

off-farm labor in each time period is

(4)\begin{equation}l({{k_t}} )= \begin{cases} 0& \textrm{if }{k_t} = H{M_t}\\ l & \textrm{if }{k_t} = H{O_t}\\ L & \textrm{if }{k_t} = A_{t} \end{cases},\end{equation}

in which HM requires the full allotment of labor time and off-farm labor hours are zero. When HO, the farmer has a fixed number, $l,$ of hours (unused maintenance hours) to contribute to off-farm work. With A, the farmer gives his entire labor allotment, L, to off-farm work. Costs also depend on the action:

(5)\begin{equation}c({{k_t}} )= \left\{ {\begin{array}{@{}ll} {{c_h} + {c_m} + {c_g} + {c_c} }&{\textrm{if }{k_t} = H{M_t}}\\ {{c_h} + {c_m} + {c_g}}&{\textrm{if }{k_t} = H{O_t}}\\ 0&{\textrm{if }{k_t} = {A_t}} \end{array}.} \right.\end{equation}

Harvest-related costs are ${c_h}$, maintenance-related costs are ${c_m}$, and any policy-related costs are ${c_g}$. These policy costs can be payments (e.g., certification costs or interest payments) that increase the farmer's operating costs or receipts (e.g., loans or PES) that reduce costs. The benchmark case contains no policy-related costs (${c_g}$ = 0). The final costs are intra-year credit costs, ${c_c},$ for short-term loans to be repaid at harvest, which reflect regional credit access described by the stakeholders. Additional constraints on the farmer include those related to the irreversibility of abandonment,

(6)\begin{equation}\textrm{if }{k_t} = A, \textrm{then }{k_{t + j}} = A\quad \forall \textrm{ }j > 0,\textrm{ }\end{equation}

to per-period subsistence constraints,

(7)\begin{equation}{v_t}({{k_t},{p_t}} )+ {W_{t }} \ge S \forall \textrm{ }t > 0,\end{equation}

and to wealth accumulation:

(8)\begin{equation}{W_t} = \left\{ {\begin{array}{@{}ll} {{W_{t - 1 }} + [{{v_{t - 1}}({{k_{t - 1}}} )- S} ]\tau }&{\textrm{if }{v_{t - 1}}({{k_{t - 1}}} )- S > 0}\\ {{W_{t - 1 }} + {v_{t - 1}}({{k_{t - 1}}} )- S}&{\textrm{if }{v_{t - 1}}({{k_{t - 1}}} )- S \le 0} \end{array}} \right..\end{equation}

In each time period, the farmer must meet a minimum subsistence income level, S, by spending down accumulated wealth, ${W_t}$, spending some of the value of the management decision in that year,$ {v_t}({{k_t},{p_t}} )$, or some combination of the two. Wealth is a state variable whose equation of motion or accumulation depends on the farmer's wealth in previous time period ${W_{t - 1 }}$, the value of the management decision last year, the subsistence level and an exogenous rate of savings, $\tau$, on income above subsistence (equation (8)).

Because coffee prices are serially correlated, the price in time t + n,

(9)\begin{equation}{p_{t + n}} = f(p_{(t-1)+n}, p_{0},p_{g},\varepsilon_{t}),\end{equation}

is a function of the price in the previous year, $p_{(t-1)+n}$; the mean price that is also the initial price, ${p_0}$; a governmental price support ${p_g}$ (e.g., price floor or conservation payment); and a random error term, ${\varepsilon _t}$ (e.g., Williams and Wright, Reference Williams and Wright1991). In the benchmark, price supports are zero $({p_g} = 0).$ In each period, the farmer forms price expectations using the known actualized price for that period, equation (9), and knowledge of the distribution of the random error term.

The value of coffee production changes across years due to price variation and as a function of the impact of prior maintenance on yields; the non-Markov characteristic of yield means that a farmer's current decisions alter the values available in future periods. Therefore, the farmer considers the expected values in T future years, based on the current year's information about those values, and uses that information to decide on the current year's action, but the farmer waits to make the decision about the next year's action until that year's price information is available (Albers, Reference Albers1996). In each period, the farmer solves equation (1) subject to per period constraints and the equations of motion in equations (2)–(8). In all, the model focuses on the joint production-labor allocation and abandonment decisions amid price uncertainty when current decisions about incurring maintenance costs determine future yield levels, and the implications of policy costs and benefits when available to the farmer.

4.1 Numerical method to solve the farmer's stochastic dynamic optimization

We develop and employ a backward induction numerical solution method in MATLAB to determine the dynamically-optimal, current period decision based on the current level of price and state variables, while facing uncertainty about future, serially correlated, and stochastically-variable prices; path dependence in yield and abandonment; and within-period constraints. The farmer considers T – here 10 – future time periods of uncertain prices and decision pathways in making his current period decision.Footnote ² The program computes the farmer's expected net present value of each current production decision based on the path-dependent expected value of every possible future production decision, conditional on meeting the within-period constraints. Then, the program identifies the highest expected net present value of the possible current period decisions – HM, HO, or A – and selects that choice as the solution to the individual period's stochastic dynamic optimization problem (see online appendix, sections V and VI).

4.2 Time path simulation framework

The individual current period decision rule cannot provide information about how per-period choices accumulate over time in the presence of path-dependent yield levels and irreversible abandonment. For this purpose, it is informative to simulate a series of constrained dynamically-optimal farmer decisions to elucidate time patterns in decisions and responses to price variability. To provide that time path information, we develop a simulation framework that defines a series of farmer optimal current period decisions. The simulation uses the numerical method described in section 4.1 to solve for the first period's optimal production decision; that production decision provides the ‘initial condition’ for yield in the second period. The second period's problem is then solved after drawing a realized price for the time period and employing the numerical method in section 4.1. The simulation continues through 20 years of the farmer making a current year optimal decision based on a single realized coffee price, all previous decisions' impact on yield, and forward-looking 10 years. We repeat the 20-year decision pathway simulation analysis for a total of 1,000 price path series iterations – randomly generating the prices in each iteration and consequently 1,000 different price paths for consideration (see online appendix, equation (A1) and section VII). Over those 1,000 price paths, we identify and explore how actualized time paths of prices interact with the path dependence of actions to influence time paths of actions when the farmer makes decisions under uncertainty about future prices. We identify a baseline percentage of these price paths in which the representative farmer abandons by each year, and the set of farmer decisions leading up to abandonment. Because we consider one representative farmer in 1,000 price path scenarios, we cannot interpret our results as a percentage of farmers abandoning by a particular time because it is the price path that varies and not the farmer who varies across these 1,000 price-outcome paths.

4.3 Parameterization to reflect the Oaxacan setting

We use the data gathered from our stakeholder interviews in Oaxaca to parameterize the model through calibration that reflects observed actions and trends. The discussions with farmers, cooperatives, extension workers and wholesalers gave us specifics on costs, wealth, wages, labor hours and yield observations (online appendix, table A1).Footnote ³ Although dated, the data provides enough information to create a stylized model characterizing the region's typical farmer. The parametrized model captures a main point emphasized by the farmers and extension workers – forgoing maintenance of coffee bushes led to a noticeable reduction in yield (Bustamante et al., Reference Bustamante, Isaza, van Heeran, Torres and Romero2009). In the absence of specific agronomic estimates of this relationship, we used reported yield and yield changes from the semi-structured interviews to calibrate parameters and functions that correspond to the observed biomass and yield growth reported by farmers and extension workers.

4.4 Policy characterizations

We introduce various policies into the representative farmer's decision to explore the impact of policies on each current period decision but also on how the pathway of those decisions evolves over time for different stochastically-defined price paths. In particular, we are interested in examining the impact of policies on farmers' decisions to forgo maintenance and to abandon coffee farming. To consider the impact of policy, we compare the no-policy decision paths to with-policy decisions paths, over 1,000 identical price paths. The policies alter the prices and/or costs (online appendix, section II); we analyze PES that impact costs, price floors that impact prices, and price premiums that can impact both prices and costs.

6.1 Price premiums

By the nature of the production process, the coffee grown in Oaxaca is organic and bird-friendly, which generates price premiums in the international market following certification. Specifically, the farmers could certify coffee as UTZ - Rain Forest Alliance, USDA Organic or Bird Friendly; however, farmers face a lengthy and costly certification process prior to qualifying to receive the price premium. We analyze policies through which farmers receive a 5 per cent (low), 15 per cent (average), and 25 per cent (high) price premium (FAO, 2009) with no, low or high certification costs, and examine different start times for the price premium implementation. Here, we assume the farmer is ‘auto-enrolled’ in the price premium program but consider a first period decision to ‘opt-out’ of the program in the case of costly certification.

Without certification costs, price premiums are quite effective at slowing abandonment (section 1 of table 1). The abandonment percentage declines by 50 per cent (low), 92 per cent (average), and 99 per cent (high) by Year 10 if the price premium policy is implemented immediately in Year 1 of the time horizon.Footnote ⁶ Even when the policy is delayed and started in Year 5, the abandonment percentage falls by 8 per cent. Premiums are an effective anti-abandonment policy for two reasons. First, they shift up the distribution of prices, which creates more profit from HM in good years and enables savings to accumulate. That wealth added to the higher price makes farmers able to perform maintenance in subsequent bad price years, avoiding periods of HO. Second, the price increase makes bounce-backs more frequent and long-lasting. Compared to the baseline's 16 per cent of price paths with bounce-backs to HM from HO, average premiums nearly double (to 32.2 per cent), and high premiums almost triple (to 44.6 per cent), the number of price paths with bounce-backs, while the number of those bounce-backs that result in abandonment declines from 42 to 5 per cent (average) and 2 per cent (high) (online appendix, table A4). The premiums make the farmer more resilient to bad price years. The bounce-backs are less likely to result in abandonment because the farmer is able to use accumulated wealth rather than forgoing maintenance. By the time the farmer must seek off-farm employment, his coffee bushes have been well maintained, making the one or two years of HO less detrimental to the operation. The farmer can then bounce-back to HM without facing low enough yields to make coffee production unprofitable because the increase in prices from the premium offsets the losses in yield.

6.2 Price premiums with certification costs

In reality, however, certification comes at significant cost to farmers in terms of time and money. Before ever receiving the premium, farmers must pay to have inspectors survey and approve the property. The costs of inspection include travel, reporting, translation and compliance fees, followed by a waiting period of three years for certification and the premium. To include certification costs in our analysis and, reflecting this setting, we assume that the farmer pays certification costs for three years before receiving the premium. Following Batz et al. (Reference Batz, Albers, Ávalos-Sartorio and Blackman2005), we chose 1,000 pesos per hectare per year for the high cost cases and 500 pesos for low cost cases. In many of the cost/premium scenarios, abandonment increases compared to the no policy case (section 2 of table 1). For the high cost, low premium case, the abandonment percentage increases by as much as 50 per cent compared to baseline, which is the opposite impact as for the no cost, low premium case. The upfront certification costs are too high for the farmer to cover without seeking off-farm employment. In addition, he must forgo maintenance earlier in more situations than without these costs. Forgoing maintenance helps him pay for certification, but by the time he receives the price premium, the resource stock has been depleted and coffee production is no longer profitable – bounce-backs are more frequent but lead to later abandonment (online appendix, table A4). With these high costs and low or average premium values, the policy can lead to even more abandonment than in the baseline (section 2 of table 1).

As a caveat, some farmers in these policy situations would decide not to become certified to avoid the high certification costs and would continue selling coffee without marketing it as shade-grown. For the policy cases in which abandonment increases compared to the baseline, we evaluated the pathways of farmers who had a first period choice to ‘opt in’ or ‘opt out’ of the policy rather than being auto-enrolled. In those cases, the farmer optimally ‘opts out’ in 24 to 32 per cent of the price path iterations, but farmers on many of those price paths still abandon.Footnote ⁷ The opt-out option reduces the abandonment percentage across cases and across time, as compared to the auto-enroll case. Despite the ability to opt out, the policy's high certification costs interact with subsistence constraints to increase the abandonment percentage during the early years of the policy, as compared to the no-policy baseline abandonment percentage. In most cases, however, the ability to opt-out of the policy enables the policy to reduce abandonment percentage by Year 20 as compared to the no-policy baseline (online appendix, table A5).

6.3 Price premiums with certification costs and loans

Because the upfront costs of certification exacerbate abandonment (or limit enrollment), we consider a hypothetical loan program to help cover the certification costs in which farmers pay back the loan with interest within three years of starting to receive the price premium. Using an interest rate of 9 per cent, which is comparable to other agricultural loan programs in Mexico, we examine the abandonment percentages when the loan covers only a portion of the certification costs (low loan), all of the certification costs (medium loan), and covers all costs plus additional support (high loan). The size of the loan and the premium determines the direction and magnitude of impact of these policies on abandonment (section 3 of table 1). A low loan combined with a small premium does not help; the abandonment percentage rises. Because the loan does not cover all certification costs, farmers must still forgo maintenance to earn enough income to pay the remaining costs. Once the farmer starts receiving the premium, the premium is not large enough to make up for yield declines from lack of maintenance and the loan payments. In order for a low loan to reduce abandonment percentages, the premium must be larger. When combining a larger loan with a small premium, more abandonment can occur because the farmer is unable to pay back the loan. The larger loan helps delay abandonment because the farmer does not forgo maintenance as early but, upon receiving the price premium, the increase in income does not cover loan payments. To fulfill the loan payments, the farmer must seek off-farm employment and the path toward abandonment begins. In general, the loan must be large enough to cover the costs of certification and the premium received must be large enough to help cover the interest and loan payments. In such cases, farmers avoid forgoing maintenance, and HO throughout the certification loan repayment periods then capture benefits of the premium.

6.4 Price floor

In the past, Mexico has used a price floor policy to help protect farmers against years of low coffee prices. The now defunct Mexican Coffee Council (CMCAFE) initiated the Fund for the Stabilization, Strengthening and Reordering of Coffee Production (Fondo de Estabilización, Fortalecimiento y Reordenamiento de la Cafeticultura) in 2001 to provide registered coffee producers with a guaranteed price (Ávalos-Sartorio and Blackman, Reference Ávalos-Sartorio and Blackman2010). Similarly, farmers are eligible for a guaranteed price if they are Fairtrade participants (Fairtrade, Reference Fairtrade2019). During the ‘Coffee Crisis’, the price floor was 675 pesos per quintal of pergamino, which is about 10 per cent lower than the mean price in our analysis. Implementing a 675 peso price floor in the first year reduces abandonment to zero in all price paths (section 1 of table 2). Slightly lower price floors of 650 and 625 pesos did not eliminate abandonment but were effective. As compared to the baseline, even the lowest price floor that was considered decreased the abandonment percentage by 34 per cent by Year 20. These price floors guarantee a future price that is enough to meet subsistence needs without off-farm wages, allowing continual maintenance to protect yield.

Table 2.

Abandonment results for price floor and PES policies

Notes: Section 1: Abandonment percentages for the baseline – no policy – model as compared to the abandonment percentages for all price floor policy runs. For the baseline run, abandonment occurs in 24% of the 1,000 price path iterations by Year 10. The percentage in which abandonment occurs grows to 35% by Year 20. When implementing a low price floor (row 1) immediately in the first period (t = 1, 1st), the percentage that abandon falls to 21% by the 10^th year (result column 1). When compared to the baseline, this represents a 13% reduction in abandonment by the 10^th year (result column 1, value in parentheses).

Section 2: Abandonment percentage for the baseline – no policy – model as compared to the abandonment percentage for all PES policy runs. For the baseline run, abandonment occurs in 24% of the 1,000 price path iterations by Year 10. The percentage in which abandonment occurs grows to 35% by Year 20. When implementing a low PES (row 1) immediately in the first period (t = 1, 1st), the percentage that abandon falls to 10% by the 10^th year (result column 1), producing a 58% reduction in abandonment by the 10^th year (result column 1, value in parentheses).

6.5 Payment for ecosystem services

Mexico's ‘National Forestry Program’ rewards farmers for maintaining forest cover to produce ecosystem services with a minimum annual payment of 280 pesos/hectare to farmers, with higher payments (medium 382 pesos/ha/yr and high 550 pesos/ha/yr) for higher levels of ecosystem services (PRONAFOR, 2014). Receiving an annual payment of 280 pesos decreases the abandonment percentage by Year 10 by 58 per cent, with larger magnitudes of the impact at higher payment levels (section 2 of table 2). This policy helps create a buffer of wealth, from savings, with which to weather bad price years. Although payments for forest cover do not create a marginal incentive for shade-grown coffee production itself, through this savings accumulation, the farmer is better protected against later low price years and can avoid forgoing maintenance and HO more often. The shade-grown coffee farmer also has the added benefit of knowing that the PES payment is certain income in each year, which lowers the price the farmer would need to observe in order to continue with HM in future periods.

6.6 Timing

The timing of the policy intervention is critical to its impact because farmers' near-term decisions to forgo maintenance affect their ability to respond to the policy later, and in our analysis, they do not know of the forthcoming policy. The efficacy of each policy drops with later implementation because some farmers have already begun the path to abandonment. If premiums, PES payments, and price floors start too late, they do not help create the buffer of wealth needed to perform maintenance during bad price years. In addition, by the time the policies are in place, some farmers are already experiencing yield declines such that even large price supports cannot always restore profitability. To examine the impact policy timing has on decisions, we analyze the five lowest price paths from our 1,000 price paths; and five paths with a significant price drop after the 4th year, focusing on average- and medium-sized policies.

For each of the lowest price paths in the baseline (figure 1, row 1), HM is chosen less than HO, and the length of time until A is relatively short. When prices are low, the farmer seeks off-farm employment and forgoes maintenance immediately. The most successful policies for reducing abandonment for all price paths are a price premium or price floor implemented in the 1st year without certification costs (figure 1, 2^nd row of both panels). These same policies are less effective if implemented in Year 5 (figure 1, 3^rd row of both panels) and even less so if the policy adds to the farmer's financial burden (figure 1, both panels, 4^th row and above). The response to delayed policies that pose excess costs do not significantly differ from the baseline (e.g., figure 1, top panel, row 1 vs row 6). Similar results occur in the set of cases with significant price drops in time period 4 (figure 2). Policies implemented in Year 1 reduce the effects of the later price drop (figure 2, top panel, row 2; and bottom panel, rows 2, 4 and 6) but policies implemented in Year 5 have less impact (e.g., figure 2, bottom panel, rows 3 and 7). To have the best chance of avoiding abandonment, the policy needs to be low cost and occur early for policies to deter the start of yield degradation from no maintenance.

Figure 1.

Decision comparisons across the lowest price paths.

The price path 1, in the baseline (column 1, row 1) the farmer harvests and maintains (HM) for the first 4 years, then harvests only (HO) for the next 4 years, and then abandons (A). In that same price path for an average premium with no certification costs starting in the 1^st year (column 1, row 2), the farmer HMs for 5 years, HOs for the next 5 years, and then abandons. A bounce indicates that the farmer shifts to HO or HM for only one period. In the baseline for price path 5, the farmer HO in Year 5 but returns to HM in Year 6

Figure 2.

Decision comparisons across price paths with a significant price drop in Year 4.

The five lowest price paths (of the 1,000 iterations) have the lowest average price for the 20-year time horizon. The figure compares the decisions over the 20 years for the baseline run of each low price path and the decisions for that same price path for select policies. Take for example, price path 1, in the baseline (column 1, row 1), the farmer harvests and maintains (HM) for the first 4 years, then harvests only (HO) for the next 4 years, and then abandons (A). In that same price path for a medium PES starting in the 1^st year (column 1, row 4), the farmer HMs for 5 years, HOs for the next 5 years, and then abandons. A bounce indicates that the farmer shifts to HO or HM for only one period. In the baseline for price path 5, the farmer HO in Year 5 but returns to HM in Year 6

6.7 Policy costs and ‘bang for the buck’

Early implementation of high value and low cost policies (to the farmer) reduce abandonment. These policies impose costs on governments, NGOs and coffee buyers, albeit while reducing the risk of biodiversity loss. To characterize the cost effectiveness, or bang for the buck, of policies, we calculate the average avoided abandonment per discounted dollar (or unit of currency) spent for each policy. We consider who bears the policy costs to determine the cost effectiveness from the perspectives of the Mexican Government and global society (table 3). Here, we consider the Mexican Government's responsibility for funding price floor, PES payment, and loan program policies,Footnote ⁸ in addition to a scenario in which they cover certification costs. When the Mexican Government does not cover certification costs, it prefers any of the price premium policies to the price floors or PES payments (table 3, columns 8 and 9) because it bears few costs for these abandonment-reducing policies. Instead, the cost burdens associated with price premium policies fall on the global society; international coffee buyers pay the premium and farmers pay for the certification. If the Mexican Government covers certification costs, the average cost of paying for certification is still lower than costs for other policies (table 3, columns 6, 7 and 11). First year price floors present high Mexican Government-borne costs but produce large reductions in abandonment. The PES and price floor policies rank toward the bottom of cost effectiveness for the Mexican Government because of the high costs it bears. Although high price floors reduce abandonment the most among policies here, the Mexican Government prefers the low floor implemented in the first year due to low costs that are paid only in low price years. PES prove costliest to the Mexican Government (table 3, columns 11 and 12) because the payments must be paid every year. Although PES effectively avoid abandonment, the Mexican Government can reach similar avoided abandonment levels with cheaper policies.

Table 3.

Policy performance and average costs

Notes: Average avoided abandonment columns (4–9): Ranking is from best (1) to worst (34). Bolded rankings represent the five best and five worst policies. A dash represents the inability to calculate a ranking for that policy (e.g., the high price floor has no abandonment, indicating there is no average amount of time that the farmer will harvest only (HO) before abandoning (A).

Average costs columns (10–13): Bolded values represent the highest and lowest average costs.

The perspective of global society includes all costs: consumers bear price premium costs, farmers incur certification process costs, costs farmers pay in interest with loans, and the Mexican Government incurs costs from PES payments and price floors. As a result, the policies with the best bang-for-the-buck for the global society are price floors and PES payments (table 3, columns 4 and 5). The price floor policies outrank all other policies because they impose the lowest cost to the society (table 3, column 13) and perform well in avoided abandonment, followed by PES policies and price premiums without loans. The least cost-effective policies are price premiums with certification costs and loans, because both farmers and consumers incur costs, and because the delayed implementation from the certification process reduces the avoided abandonment. From both the Mexican Government and Global perspectives, savings from delaying implementation do not outweigh the extra abandonment.