Costs and benefits of reducing non-point pollution from farming

1
Costs and benefits of reducing non-point
pollution from farming

• Nick Hanley
• Economics Department
• University of Stirling, Scotland

2
outline

• Policy context Water Framework Directive,
  Nitrates Directive
• Estimating costs of pollution control focus on
  transferability of policy option rankings between
  two catchments
• Estimating benefits of pollution control uses
  choice experiment method, and again looks at
  transferability of benefit estimates between the
  two catchments.

3
The WFD

• The Water Framework Directive (2000/60) contains
  a number of ambitious aims for the future of
  water resource management in the EU. These
  include
• protecting and enhancing aquatic ecosystems and
  wetlands
• promoting the sustainable long-term use of water
  resources
• progressively reducing emissions to the water
  environment
• putting incentives in place to encourage users to
  use water resources efficiently and
• contributing to the mitigation of flooding and
  droughts.

4

• Several key principles underlie these aims in the
  Directive, including implementation of the
  polluter pays principle, management of rivers on
  a river basin basis, and the setting up of
  cost-effective plans to achieve Good Ecological
  Status (GES) in all EU waters (except for cases
  of disproportionate costs).
• The Water Framework Directive requires Member
  States to put in place Programmes of Measures
  (PoMs), made operational through the
  implementation of three iterations of River Basin
  Management Plans starting in 2009 and ending in
  2027. The Directive requires Member States to
  select measures on the basis of environmental,
  economic and social criteria, with the aim of
  achieving the most cost-effective combination of
  measures..
• And then assessing their costs and benefits to
  determine and justify exemptions.

5

• Will what is cost-effective as a way of tackling
  a given water quality problem in one catchment be
  cost-effective in another?
• How transferable are policy options in terms of
  their aggregate costs?
• How transferable are the benefits?
• We examine this, looking at 2 linked water
  quality problems, both associated with farming
• Low summer water levels (ecological problems due
  to oxygen deficits)
• High nutrient levels (ecological problems due to
  eutrophication and consequent algal blooms)

6
Previous work

• Much existing work on the relative costs of
  economic instruments and managerial measures to
  control Non Point Source (NPS) pollution from
  farming
• Most only looks at one water body
• Most only looks at one environmental problem
• Some papers do link NPS pollution with irrigation
  demand, but no explicit link to maintaining river
  flows

7
methodology

• Selected two similar catchments, each with a
  problem in attaining Good Ecological Status due
  to low flows and nutrient levels. Farming
  implicated in both cases as main source of
  problem.
• Construct BPE (BioPhysical Economic) models for
  each catchment, linking land use, water quality
  and water quantity (flows). Treats whole of each
  catchment as one farm. Based on 10 years of
  economic/environmental data. Calibrated to 2001.
• Simulate a range of policy options
• Rank in terms of aggregate cost

8
Economic model

• Whole catchment as single farm, maximising
  profits subject to resource constraints and
  environmental constraints
• Complex non-linear system
• Range of cropping activities, with associated
  revenues, costs and rotation constraints
• Heterogeneity via different soil types
• Can simulate economic instruments via objective
  function
• Can also simulate changes in CAP

9
Environmental Components

• Include equations which summarise outputs from an
  off-line water quality model which relate soil
  type, rainfall, land use, fertilizer use and
  animal numbers to nitrate levels in river. Allows
  us to calculate shadow prices for different
  environmental constraints set in terms of nitrate
  concentrations
• Include constraints relating use of irrigation
  water to water availability in the catchment over
  a 10-year period, so we can also back out a
  shadow price for scarce irrigation water inputs
  to farming
• Use (1) and (2) to constrain model to hit
  variable environmental targets in terms of
  minimum summer flows and maximum N03 levels

10
The Motray
The Brothock
11
Results costs of achieving targets

• Depend on degree of flow restriction (water flows
  target)
• Depend on severity of water quality target
  (ambient standard exceeded 10 or 5 of the time
  over a year)
• Depend on what kind of set-aside is allowed
  (permanent does best in terms of reducing N
  run-off)

12
Motray ranking of policy options in terms of
resource costs
Regulatory Target Standard exceeded 10 of time Standard exceeded 10 of time Standard exceeded 10 of time Standard exceeded 5 of the time Standard exceeded 5 of the time Standard exceeded 5 of the time
River Flow Restriction No Flow restriction 95th percentile 90th percentile No Flow restriction 95th percentile 90th percentile
Input Tax 1 1 1 2 2 2
Input Quota 2 2 2 3 3 3
Stocking density input tax 6 6 6 5 5 5
Set-aside Input tax 3 3 3 1 1 1
Set-aside stock density 5 5 5 4 4 4
Set-aside 4 4 4 6 6 6
Stocking density 7 7 7 7 7 7
So pure economic instrument does best at lower
standard, but mixed instrument better under
higher standard
13
Results for Brothock

• Ranking of options is identical to Motray
• But instrument levels differ eg required N tax
  higher in Motray under all scenarios (eg 45
  rather than 41 eg quota is -36 rather than
  -31).

14
Absolute levels of instruments in Brothock with
no restriction on type of setaside
Regulatory Target Standard exceeded 10 of time Standard exceeded 10 of time Standard exceeded 10 of time Standard exceeded 5 of the time Standard exceeded 5 of the time Standard exceeded 5 of the time
River Flow Restriction No Flow restriction 95th percentile 90th percentile No Flow restriction 95th percentile 90th percentile
Input Tax ( increase) 41 41 39 58 57 55
Input Quota ( reduction) -31 -31 -29 -37 -35 -34
Stocking density (1.4 glu/ha) input tax ( increase) 26 25 23 29 29 25
Set-aside (500 ha) Input tax ( increase) 32 32 30 36 35 33
Set-aside (500 ha) stock density ( decrease) -27 -25 -25 -32 -30 -29
Set-aside ( increase) 292 287 280 365 352 320
Stocking density ( decrease) -40 -39 -37 -51 -51 -49
15
Conclusions on costs

• Costs depend on severity of target, and whether
  aiming for joint targets (flows and quality).
• Economic instruments typically cost-effective,
  but under some circumstances (higher quality
  target) a MIX of economic instruments and
  regulation is most cost-effective.
• Note that we do not model changes in the price of
  irrigation farmers in Scotland currently pay no
  fee for water use.
• Optimal input tax varies across catchments
• But ranking of policy options is the same

16
Estimating the benefits of improvements in water
flows and water quality

• Focus on same two catchments
• Focus on same water quality and flow issues
• Question can we transfer the benefit values
  between these catchments?
• This is interesting because, under the WFD,
  environmental agencies will have to undertake a
  great many benefit transfer exercises to decide
  which improvements to Good Ecological Status are
  dis-proportionately costly
• We use a Choice Experiment to do this.
• Hanley et al, Euro. Rev. Ag. Econ., 2006.

17
Choice Experiments (CE)

• Based on characteristics theory of value and
  random utility theory
• Assumes utility function can be de-composed into
  deterministic and random components
• Train (1998) introduced the Random Parameter
  version of the model, which improves on the
  more usual conditional logit by allowing for
  preference heterogeneity estimate both a mean
  effect of an attribute on choice and the standard
  deviation of this effect. Also allows for
  correlation across choices by an individual
• Standard RPL assumes attributes are uncorrelated.
  We relax this people who like attribute a
  more might also like attribute b more.

18
RPL basic specification

• Ujn Aj ?k ?jk Xjkn ?m ?m Smn ?k ?kn Xjkn
  ?jn
• Aj is an alternative specific constant, Xjkn is
  the kth attribute value of the alternative j ßjk
  is the coefficient associated to the kth
  attribute, Smn is the mth socio-economic
  characteristic of individual n, ?m is the
  coefficient associated with the m individual
  socio-economic characteristic, ?kn is a vector of
  K deviation parameters which represents the
  individuals tastes relative to the average (?)
  and ?jn is an unobserved random term that is
  independent of the other terms in the equation,
  and which is identically and independently Gumbel
  distributed. The coefficient vector ?jk varies
  among the population with density f(??), where ?
  is a vector of the true parameters of the taste
  distribution.

19
Steps in CE design

• Choose attributes and levels
• Design choice sets
• Choose population to sample
• First two are based on some likely policy
  scenarios, described in terms of changes in
  abstraction licensing and controls on fertiliser
  use, and how these would impact on the appearance
  and ecological quality of the river
• Levels/attributes
• Ecological condition (worse, slight improvement,
  big improvement)
• Flow rate (months of low flow)
• Agricultural jobs
• Cost of programme to local households

20
(No Transcript)
21
? Low Flow
Normal Flow ?
22
Example choice card (everyone got 4 of these)
Note do nothing option is constant across all
choice sets, and corresponds to worsening of all
environmental/social attributes
23
Sampling

• Sample frame is local residents in the 2
  catchments
• Used mail shot
• Response rate 30
• Useable surveys 348 in Motray, 344 in Brothock
• Status quo chosen 10 of time
• 90 of respondents were WTP for improvements in
  local water quality

24
Results

• Two RPL models one with correlation between
  attributes, one without. Estimate separately for
  each catchment, then pool. (note Conditional
  Logit fails IIA test even with socio-economics
  in)
• Tried including socio-economic variables and
  attitude measures as well, but only attitudes
  were significant, and these are not much use for
  BT, plus meant we lost a lot of observations due
  to missing data.
• ASC was insignificant
• Note that scale (error variance) might differ
  between samples, so tested for this. Relative
  scale ratio is 0.95, Brothock has slightly lower
  response variability than Motray. Comparisons of
  model parameters allow for this.

25
RPL with corr. attributes Motray (n348) Motray (n348) Motray (n348) Brothock (n344) Brothock (n344) Brothock (n344) Join (n 692) Join (n 692)
Coeff. Std errors Std errors Coeff. Std errors Std errors Coeff. Std errors
Mean effects Mean effects Mean effects Mean effects Mean effects Mean effects Mean effects Mean effects Mean effects
Local Farm Jobs 0.581 0.581 0.12 0.467 0.467 0.11 0.528 0.083
Flow -0.813 -0.813 0.26 -0.406 -0.406 0.24 -0.543 0.148
Ecology level 1 2.364 2.364 0.94 2.219 2.219 0.99 1.570 0.355
Ecology level 2 5.143 5.143 1.25 4.572 4.572 1.21 3.982 0.575
Tax -0.217 -0.217 0.03 -0.127 -0.127 0.03 -0.155 0.021
Standard deviation terms Standard deviation terms Standard deviation terms Standard deviation terms Standard deviation terms Standard deviation terms Standard deviation terms Standard deviation terms Standard deviation terms
Jobs 0.358 0.358 0.14 0.308 0.121 0.121 0.415 0.095
Flow 0.772 0.772 0.34 0.425 0.303 0.303 0.614 0.204
Ecology 1 2.536 2.536 0.74 2.146 0.782 0.782 1.645 0.329
Ecology 2 4.334 4.334 0.79 2.522 1.347 1.347 2.472 0.560
Log Likelihood (pseudo-R2) -220.67 (0.38) -220.67 (0.38) -220.67 (0.38) -242.76 (0.34) -242.76 (0.34) -242.76 (0.34) -467.00 (0.36) -467.00 (0.36)
26
BT tests

• Are the models the same? Likelihood ratio test
  says we cannot reject the null hypothesis of
  parameter equality once we allow for difference
  in relative scale
• ß(Motray) ß (Brothock) unusual!

27
BT testing (2)

• Are the implicit prices different? Implicit price
  for attribute a ßa/ßcost.
• Test for implicit price (low flows, Brothock)
  implicit price (low flows, Motray) ,and for
  ecological quality, using Poe et al (1994) test.
  Results depends on whether use correlated
  attribute version of model or not.
• With correlation no differences in implicit
  prices for any attribute
• Without correlation jobs and big improvement in
  ecology are significantly different
• Comparing the pooled model, which might be the
  benefits transfer system, with the two catchment
  models, also get transferable estimates with the
  un-correlated attributes version of the model,
  and signif. diff. for one attribute for each
  river with correlated preferences

28
BT tests welfare estimates

• Look at three hypothetical scenarios for
  improvements in water quality
• Calculate the compensating surplus using

29
Welfare Measures for three policy scenarios
(/household/year)
WTP 95 ci WTP WTP 95 ci WTP 95 ci
Motray Motray Brothock Brothock Brothock Pooled Pooled
Scenario 1 add. Jobs0, flow 3, ecology slight improvement Ind. coef. 56.8 45.8-67.9 62.0 44.0-83.9 44.0-83.9 58.1 48.6-68.4
Scenario 1 add. Jobs0, flow 3, ecology slight improvement Corr coef. 58.3 33.8-79.1 85.0 43.3-133.6 43.3-133.6 59.7 47.24-72.2
Scenario 2 jobs 2, flow 2, ecology slight improvement Ind. coef. 67.7 55.3-80.5 72.0 53.6-93.3 53.6-93.3 68.4 57.9-79.5
Scenario 2 jobs 2, flow 2, ecology slight improvement Corr coef. 67.4 42.1-88.4 95.6 52.8-144.6 52.8-144.6 70.0 56.5-83.1
Scenario 3 jobs 5, flow 1, ecologybig improvement Ind. coef. 97.2 79.7-115.5 103.3 80.9-133.3 80.9-133.3 98.9 85.1-114.7
Scenario 3 jobs 5, flow 1, ecologybig improvement Corr coef. 91.9 65.8-113.7 128.5 85.22-179.3 85.22-179.3 99.2 83.2-11.8
V0 Base jobs -2, flow 5, ecology worsening V0 Base jobs -2, flow 5, ecology worsening V0 Base jobs -2, flow 5, ecology worsening V0 Base jobs -2, flow 5, ecology worsening V0 Base jobs -2, flow 5, ecology worsening V0 Base jobs -2, flow 5, ecology worsening V0 Base jobs -2, flow 5, ecology worsening V0 Base jobs -2, flow 5, ecology worsening V0 Base jobs -2, flow 5, ecology worsening
30

• Welfare estimates are more precise for the
  without correlation model
• Same improvements valued more highly in the
  Brothock than the Motray, although the difference
  is not significant using the Poe et al (1994)
  test
• Might make sense current water quality is
  slightly worse in the Brothock
• But in both models, using either catchment to
  predict values of water quality improvement in
  the other would not produce significant errors
  encouraging finding?

31
Conclusions - benefits

• Much greater need for benefits transfer now that
  WFD is being implemented, and river basin
  management plans are being drawn up
• Choice experiments offer quite a bit of
  flexibility as the basis of a BT system, but not
  much evidence to date as to their performance in
  this regard
• Our findings show that whether one allows for
  correlation between attributes seems to make a
  difference. Our earlier work showed that allowing
  for preference heterogeneity via RPL could reduce
  transfer errors

32

• So we are getting nearer an acceptable system?
  Seems to depend, in our paper, on how one tests
  for differences
• But what do we conclude if we accept value
  function and welfare estimate transfer, but
  reject implicit price transfer?? Depends what we
  want to know
• How close is close enough for policy purposes?
  95 is probably too strict?