The marginal effect of climate change on profits in the agricultural sector

(1)

University of Amsterdam

Faculty of Economics and Business

Bachelor’s Thesis

The marginal effect of climate change on

profits in the agricultural sector

August 2013

Author: Tom Swinkels Student ID: 10025383

(2)

Table of contents

1. Introduction……… 2. Literature review……… 3. Data and model specification

3.0. The data.……….……… 3.1. Agricultural……….………… 3.2. Weather………..

3.3. Extra control variables………. 3.3.1. Year times regional fixed effects…………....………. 3.3.2. Soil………..

3.3.3. Extra Geographic………. 3.4. Model specification………

4. Results and discussion

4.1. Results……….

4.1.1. Regression 1 with control for elevation……… 4.1.2. Regression 2 w.c.f. elevation and weather variability.... 4.1.3. Regression 3 w.c.f. elevation and elevation times temp.. 4.1.4. Regression 4 w.c.f. all extra geographic variables……… 4.2. Comparison with DG (2012)………. 4.3. Discussion……….. 5. Conclusions……… 6. References……… 2 3 6 6 6 7 7 8 8 10 15 15 19 22 25 28 29 32 35 1

(3)

1. Introduction

Climate change is a widely discussed problem. Yet, while there is still no general consensus regarding the causes and implications of climate change, the intergovernmental panel on climate change (IPCC) records do suggest that our weather is changing (McKibbin & Wilcoxen, 2002, p. 111). This paper will focus on the effects climate change might have on the agricultural sector.

Previous research on this topic by Deschenes and Greenstone (henceforth DG) (2007;2012) and Fischer et al. (2012) use USA climate and agricultural data in order to estimate the effects of climate change on agricultural profits. This paper will attempt to improve the estimates in table 1 of DG (2012) by incorporating elevation and the standard deviations of several variables in the regression analysis. More specifically, this paper will add to the estimation of the marginal effects of climate in change by adding: elevation, standard deviation of elevation, standard deviation of temperature and the control variable elevation times temperature to the analysis. Very few academic journals comment on the link between climate change and elevation and DG (2012) suggest that the omitting of elevation control affects temperature predictions in specific climate change models. Hence, elevation is added as a control variable following the assumption that climate change will have different effects at different altitudes.

Secondly, the standard deviation of precipitation controls for the variability in rainfall in county across years whereas the standard deviation in temperature controls for the variability in temperatures between years within one specific county. For example, the standard deviation of temperature or precipitation in the county of Baldwin in 1992 controls for the variability in temperature in the county of Baldwin from 1987 up to 1992. Lastly, the variable elevation times temperature controls for the interaction between elevation and temperature. Furthermore, the regression techniques are adjusted to absorb state fixed effects rather than the county fixed effects that DG (2012) use in order to reduce collinearity. With these modifications, this paper will provide a

(4)

more thorough insight on the effect of climate change across regions with different levels of elevation.

The structure is as follows. First, in section 2, a literature review is presented in which previous literature on the topic is discussed. Secondly, in part 3, a description of the data is provided. Additionally, part 3.4 elaborates on the model specification and how the estimates are calculated. Subsequently, in part 4, all the results are discussed per table. Afterwards, in part 4.2 these results are compared with the output of DG (2012). Section 4 ends with a discussion of the shortcomings of this paper. Finally, a general conclusion is presented in part 5.

2. Literature Review

The concept of incorporating elevation in climate change models is not entirely new. Previous literature by Beniston et al. (1997), Frauenfeld et al. (2005) and You et al. (2008) sheds some light on the issue. Subsequently, research by DG (2007), Fischer et al. (2012 and DG (2012) explore the effect of climate change on the agricultural sector.

Firstly, in their paper Beniston et al. (1997) investigate the effect of climate change in mountainous regions and stress the relevance of elevation and weather variability in climate change models. In their paper Beniston et al. (1997) use the Swiss Alps as a case study. They argue that mountains stimulate cloud formation and atmospheric flows through which temperature and precipitation are affected (p. 234-235). Furthermore, their research suggests that extreme climatological events, thus changes in upper and lower weather extremes induced by climate change, have a greater impact than mere changes in the mean climate (p. 238). To summarize, Beniston et al. (1997) conclude that temperature and precipitation are affected by mountains and elevation. Consequently, the effects of climate change will be felt differently in highly elevated areas. The observed pattern suggests that as elevation increases, the effects of climate change become more perceptible (p. 234-235). Ultimately, these enhanced changes could severely disrupt downstream agriculture, industry and residents that

(5)

rely on upstream resources (Beniston et al., 1997, p. 246-247). Secondly, Beniston et al. (1997) note that shifts in upper and lower weather extremes – which are induced by climate change – are likely to a more significant impact than changes in mean weather (p.238).

Frauenfeld et al. (2005) focus on weather variability in the Tibetan Plateau - the world’s most elevated plateau which is argued to have a significant impact on climate in Asia (p. 1). They stress that more climate data on high altitude regions is needed to further investigate the implications of climate change. Corresponding to the findings of Beniston et al. (1997), the Tibetan Plateau has been subject to a large increase in temperature over the last decades and according to Frauenfeld et al. (2005), this change in climate could alter the circumstances in the rest of Asia (p.1).

You et al. (2008) confirm the increased sensitivity to climate change in elevated regions in the Tibetan Plateau, however the authors point out that the relationship between elevation and climate change is not homogenous in all regions. For example, in the Andes mountains of South America, a different trend between climate change and elevation has been observed. In the Western part of the Andes, the lower elevated regions have been more sensitive to climate change than the more highly elevated areas (p.1).

From the contrast in findings by Frauenfeld et al. (2005) and You et al. (2008), we can ascertain that the effect of elevation on climate change is ambiguous in nature and requires more research.

When looking at DG (2007)’s paper titled: ‘The Economic Impacts of Climate Change: Evidence from Agricultural Profits and Random Fluctuations in Weather’ is a well-cited paper that uses USA data in six periods:1978, 1982, 1987, 1992, 1997 and 2002 to estimate the effect of climate change on profits in the agricultural sector. DG state that climate change includes changes in temperature and precipitation as is confirmed by Beniston et al. (1997) & Frauenfeld et al.( 2005). As temperature and precipitation are inputs in the agricultural sector, a change in these factors will lead to a change in agricultural outputs. Additionally, they point out that previous attempts to estimate the effects of

(6)

climate change using the cross-sectional hedonic approach are incorrectly specified. They continue by opting for a model which uses annual variation in weather to estimate the effect of climate change on profits in the agricultural sector. In order to calculate these effects on the long-run, the Hadley 2 climate change model is used to predict weather variables. Ultimately, the paper estimates that annual agricultural profits will increase by 4% by the end of the 21st_{century due to climate} change (DG, 2007, pg. 1)

In their comment on DG (2007), Fisher et al. (2012) point out several errors in the data and method. They argue that the effect of climate change on farm profits is likely to be negative and that DG’s (2007) estimates are positively biased due to: i) inaccuracies in weather data, ii) the use of the outdated Hadley 2 model and iii) the omission of storage possibilities for farmers (p. 10-11). With these corrections Fisher et al. (2012) find a negative impact of climate change in all but one specification (pg. 3750).

DG (2012) is the paper that DG provide as a reply to the criticism of Fisher et al. (2012). In this paper DG correct their data sample accordingly by removing inaccurate data and using a more

contemporary climate change model. They conclude that the estimated effect of climate change on agricultural profits are – in contrast to DG (2007) – are in fact negative. Yet, DG state that strong evidence of a negative effect is still missing (pg. 3772). Additionally, as mentioned before, DG (2012) already hint at the relevance of controlling for elevation in order to compensate biased estimates in their climatic models (p. 3764). Ultimately, DG stress that much uncertainty remains and that more research on the economic impacts of climate change is required (p. 3771-3772).

This paper will seek to continue the discussion DG (2007), Fisher et al. (2012) and DG (2012) have started and attempt to incorporate the findings of Beniston et al. (1997),Frauenfeld et al. (2005) and You et al. (2008). In other words, this paper will contribute to the estimation of the effects of climate change on agricultural profits by including elevation and weather variability variables.

(7)

3. Data and model specification 3.0 Data

This paper uses USA data at county level. More specifically, the data set consists of information on 2341 counties over a period from 1987 up to 2002. In essence, for each county, data is available in 1987, 1992, 1997 and 2002. This amounts to a total of 9364 observations that cover 84 percent of all USA farmland (DG, 2012, pg.3764). The data can be divided into three categories: agriculture-related, weather-related, and other control variables (see table 1). Each category will be discussed in more detail below.

3.1 Agricultural variables

Derived profits per acre (y) is the only variable in the agricultural category. The unit of measurement is constant 2002 US dollars ($2002). The data originate from the DG (2012) paper where it is

calculated by taking farm revenues minus farm costs and dividing by acres of farmland for each county in the USA. As section 3.4 will indicate, y is taken as dependent variable in all regressions.

3.2 Weather variables

Firstly, the weather category only includes temperature and precipitation variables. In this paper, the unit of measurement for temperature is cumulative growing season degree days defined with a base of 46.4° F and ceiling of 89.6°F. Essentially, a degree day is calculated in two steps. First of all, for each day the average temperature in a county is calculated by taking the sum of the minimum and maximum daily temperature and dividing the sum by two (in this case the maximum

temperature cannot exceed 89.6°F). Secondly, the base temperature is subtracted from the

calculated average temperature. In this case, the base temperature is 46.4° F so one would subtract 46.4 from the average temperature of each day in order to calculate the ‘degree days’ for the particular day. In order to calculate the growing season degree days, one simply needs to take the sum of all the degree day values in the growing season which in this paper ranges from April to

(8)

September. The variable name for temperature in the regressions is ‘dd89’ which essentially is an abbreviation of degree days with a ceiling of 89.6°F.

Secondly, the unit of measurement of growing season precipitation is inches which takes the total precipitation for the growing season (April to September). All weather data originates from the DG (2012) data set where the variable name is ‘prcp’. In DG (2012), the temperature data is aggregated by an inverse-distance weighted average of all the valid measurements from stations that are located within a 200 km radius of each county’s center (p. 3763). In other words, the data is adjusted in order to accurately obtain precipitation values at county level. The precipitation variables originate from the Independent Slopes Model that use a parameter-elevation regression (DG, 2012, p. 3763).

Thirdly, a distinction is made between non-irrigated and irrigated counties. A county is considered to be irrigated if at least 10% of its farmland is irrigated (DG, 2012, p. 3765). The total number of non-irrigated counties in this data set is 7743 while the number of non-irrigated counties numbers 1621. All irrigated county related weather variables have the prefix ‘irr’ while all the non-irrigated county related weather variables have the prefix ‘dry’.

Furthermore, in order to allow for a non-linear relationship between the weather variables and profits the square of temperature and precipitation are included.

3.3 Extra control variables

This section consists of three sub-sections: i) yearly times regional effects, ii) soil characteristics and iii) extra geographic variables. These sub-sections are discussed in more detail below.

3.3.1. Yearly times regional effects

This category comprises of four different effects: i) yearly, ii) year*USDA region, iii) year*census division, iv) year*state (where * denotes multiplication). The purpose of these variables is to absorb any temporary effects that may have occurred in particular years or regions. For example, if yearly

(9)

effects are included in the regression the variable will keep the yearly effects constant over all counties. By including year*USDA region, year*census division or year*state division, yearly fixed effects per USDA region, census division or state are taken, respectively for each area. The coding of these variables originates from the DG (2012) set. Essentially, a unique code for each year in a relevant region are generated.

3.3.2. Soil variables

All nine soil variables originate from the DG (2012) data set. The names of these soil characteristics vary from x1 to x9 and include: salinity, percentage of flood-prone land, percentage of wetland, k-factor (soil loss k-factor), slope length, percentage of sand, percentage of clay, moisture capacity and permeability, respectively (see table 1). The soil variables control for changes in soil quality that could cause bias estimates of the weather variables.

3.3.3. Extra geographic variables

All elevation data originate from a digital elevation map made available by the U.S. Geological Survey's Center for Earth Resources Observation and Science (EROS, 2010).

The digital elevation data have been edited in GIS to provide the average elevation per county (in meters above sea level) and is merged with the existing agricultural and climate data from DG (2012). The average elevation per county is assumed to be time-invariant so all elevation values per county are constant over the four periods of time (see figure 1. for an overview of USA elevation).

(10)

Table 1.Summary Statistics Variable

Category Variable Description Source (St.dev) Mean

Agricultural Y Profits per acre of farmland ($2002) DG (2012) 65.56 _(114.48)

W

ea

th

er

Dry Indicator if less than 10% of farmland is irrigated DG (2012) 0.83 (0.38) Dd89 Cumulative growing season degree-days, defined

with a base of 46.4°F and ceiling of 89.6°F DG (2012) 3964.76 (1069.49) Prcp Growing season precipitation (inches) DG (2012) 19.75

(7.55) Dry_dd89 Cumulative growing season degree-days, defined

with a base of 46.4°F and ceiling of 89.6°F in non-irrigated counties

DG (2012) 3284.23 (1781.16) Dry_dd89_sq Cumulative growing season degree-days, defined

with a base of 46.4°F and ceiling of 89.6°F in non-irrigated counties squared

DG (2012) 1.40E7 (1.00E7) Dry_prcp Growing season precipitation (inches) in

non-irrigated counties DG (2012) 17.11 (9.91) Dry_prcp_sq Growing season precipitation (inches) in

non-irrigated counties squared DG (2012) 390.99 (307.49) Irr_dd89_sq Cumulative growing season degree-days, defined

with a base of 46.4°F and ceiling of 89.6°F in irrigated counties squared

DG (2012) 2904624 (7479462) Irr_prcp Growing season precipitation (inches) in irrigated

counties DG (2012) 2.64 (7.00)

Irr_prcp_sq Growing season precipitation (inches) in irrigated

counties squared DG (2012) 55.96 (192.71) m_dry_dd89 Mean of variable dry_dd89 with analytic weights

using farmland DG (2012) = 3064.76 (1805.65) m_irr_dd89 Mean of variable irr_dd89 with analytic weights

using farmland DG (2012) = 756.27 (1616.02) m_dry_prcp Mean of variable dry_prcp with analytic weights

using farmland DG (2012) = 13.31 (9.23) m_irr_prcp Mean of variable irr_prcp with analytic weights

using farmland DG (2012) =2.50 (6.50) Ot he r c on tr ol

Iyear* Census of agricultural year DG (2012) - Iuuyy* Interaction between year and USDA region

variable DG (2012) -

Iddyy* Interaction between year and US Census division DG (2012) - Issyy* Interaction between year and state DG (2012) -

X1 Salinity DG (2012) 0.02 (0.05) X2 Fraction flood-prone DG (2012) 0.15 (0.21) X3 Wetlands DG (2012) 0.10 (0.12) X4 K-factor DG (2012) 0.29 (0.09) X5 Slope length DG (2012) 214.42 (167.50) X6 Fraction sand DG (2012) 0.09 (0.22) X7 Fraction clay DG (2012) 0.18 (0.250) X8 Moisture capacity DG (2012) 0.17 (0.04) X9 Permeability DG (2012) 2.891 (2.86) Elevation Average county elevation (meters) EROS (2012) 488.58

(534.94)

(11)

3.4 Model specification

The aim of this paper is to estimate the marginal effects of a change in weather on profits per acre. In short, the process involves several regressions using profits (y) as dependent variable and the

weather and extra control variables as independent variables. From these regressions, the

coefficients of the growing season temperature and precipitation variables are taken and inserted into a linear equation that calculates the marginal effects of a change in weather on agricultural profits. This section will elaborate further on the process of this model. In order to make the model more coherent, the structure is divided into three steps.

Firstly, as mentioned in section 3.2, a distinction is made between irrigated and non-irrigated counties. To reiterate, an irrigated county is considered to be a county of which at least 10% of its farmland is irrigated. In order to make the distinction, four new weather variables are created that are allowed to vary between the irrigated and irrigated counties. These new variables are: non-irrigated temperature, non-non-irrigated precipitation, non-irrigated temperature and non-irrigated precipitation. Where non-irrigated temperature represents the temperature in non-irrigated counties and where non-irrigated precipitation signifies precipitation in non-irrigated counties and so on (see table 1). The regression coefficients of these four weather variables are used in a later stage to form the basis of the final estimates of the marginal effect of climate change.

The second step consists of four sets of four regressions that use agricultural profits (y) (in $2002) as dependent variable. The type of regression that is used in step two is a fixed effect regression analysis that: i) uses area of farmland in each county as analytic weight, ii) distinguishes between

Ot

he

r c

on

tr

ol Stdevdd89 Standard deviation of dd89 variable across years DG (2012) 264.27 _(191.087)

Stdevprcp Standard deviation of prcp variable across years DG (2012) 2.74 (2.21) Elevation*dd89 Interaction between elevation and dd89 variable EROS (2012),

DG(2012) 1720318.00 (1762661.00)

(12)

county and state clustered standard errors and iii) absorbs for state fixed effects. The regressions are consistent with the subsequent structure:

Y = α + X'β + Z'γ + ε

Where X and Z are vectors of weather and other control variables, respectively. In addition, β and γ are the coefficients of the respective variable vectors and α is a constant. Lastly, ε portrays the error term.

As mentioned above, the sets of regressions are fourfold. In each set the weather vector (X) remain unchanged. In other words, all regressions use the same weather variables (all listed in table 1). Of the other control variables (Z), all nine soil characteristics and yearly times regional fixed effects are included as independent variables in all regressions. Between the four sets of regressions, the only difference lies in the extra geographic sub-section of the Z vector. Thus, the sub-section that includes the following variables: elevation, standard deviation of elevation, standard deviation of temperature and elevation times temperature. To elaborate, the difference between the regressions lies in the choice of extra geographic variables. In order to clarify the exact differences, the exact composition of the four regressions are displayed below.

1. The first regression consists of all the variables in the X vector as well as the following variables for the Z vector: all soil characteristics (x1 up to x9), all year times regional effects (iyear* up to issyy*) and elevation.

2. The second regression consists of all the variables in the X vector as well as the following variables for the Z vector: all soil characteristics (x1 up to x9), all year times regional effects (iyear* up to issyy*), elevation and the standard deviations of temperature (stdevdd89) and precipitation (stdevprcp) in the other control category .

3. Subsequently, the third regression consists of all the variables in the X vector as well as the following variables for the Z vector: all soil characteristics (x1 up to x9), all year times

(13)

regional effects (iyear* up to issyy*), elevation and elevation times temperature (elevation*dd89).

4. Finally, the last regression consists of all the variables in the X vector and all the variables in the Z vector.

For each of the four sets above, four regressions are carried out – each with a different yearly times regional fixed effect. To clarify, in the case of regression 1, the regression is carried out four times (1.1 up to 1.4) as described above. Regression 1.1 only includes yearly fixed effects (iyear*), regression 1.2 only includes yearly times USDA region fixed effects (iuuyy*), regression 1.3 only includes yearly times Census division fixed effects (iddyy*) and regression 1.4 only includes yearly times state fixed (issyy*).

The structure of the yearly times regional fixed effects is the same as in DG (2012). In tables 3-6, column A shows the regression with yearly effects, column with yearly times USDA regional effects, column C with yearly times census division effects and column D with yearly times state effects.

In the third step of the process, estimates on the marginal effects of climate change are calculated by taking a linear combination of four relevant weather coefficients. Essentially, for the effect of temperature, the β for temperature in the dry/irrigated county is added with 2 times the mean temperature for that county times the β for temperature squared. This gives the marginal effect of an increase of one growing season degree day on agricultural profits. As an increase of one growing season degree is negligibly small, the figure is multiplied by 100 to symbolize an increase of 100 growing season degree days. In equation form, the general structure for the estimates of an increase in temperature are:

100*(‘ β temperature in irrigated/irrigated counties’ + 2*`mean of temperature in non-irrigated/irrigated counties'*’ β temperature in non-non-irrigated/irrigated counties squared’)

(14)

For precipitation, the β for precipitation in the dry/irrigated county is added with 2 times the temperature mean in dry/irrigated county times the β for precipitation squared in dry/irrigated county. In equation form, the general structure for the effects of an increase of precipitation by 1 inch is:

‘β precipitation in irrigated counties/irrigated’ +2*`mean of precipitation in non-irrigated/irrigated counties '* ‘β precipitation in non-non-irrigated/irrigated counties squared’ The full set of these linear combinations are displayed below in variable form. These linear combinations apply for all the estimates displayed in tables 3b, 4b, 5b and 6b.

For the marginal effect of growing season degree days on agricultural profits in non-irrigated counties (in row 1, columns 1a-1d) , the following equations of coefficients (β) are applied :

100*(‘ βdry_dd89’+ 2*`m_dry_dd89’*’βdry_dd89_sq’)

For the effect of growing season degree days in irrigated counties on agriculture (in row 1, columns 2a-2d) the equation of coefficients is as follows:

100*(‘ βirr_dd89’+ 2*`m_irr_dd89’*’βirr_dd89_sq’)

Subsequently, for the effect of growing season precipitation in non-irrigated counties (in row 2 columns 1a-1d) the relevant equation is:

(‘ βdry_prcp’+ 2*`m_dry_prcp’*’βdry_prcp_sq’)

Lastly, in irrigated counties the linear combination of weather coefficients for growing season precipitation is (in row 2 columns 2a-2d):

(‘ βirr_prcp’+ 2*`m_irr_prcp’*’βirr_prcp_sq’)

(15)

The structure of the output of the equations described above is displayed in table 2 below. Note that in table 2, the equation is repeated for each yearly times regional fixed effect. Thus in row 1 and 2, the equation described will be applied four times for non-irrigated counties and four times for irrigated counties in columns A, B, C and D.

Table 2. Calculation of estimates for tables 3b, 4b, 5b, 6b

Non-irrigated Counties (N=7743) Irrigated Counties (N=1621) Marginal effects

(at sample mean) 1a 1b 1c 1d 2a 2b 2c 2d

1. Growing season degree

days x100 2*`m_dry_dd89’*’β100*(‘ βdry_dd89dry_dd89_sq’+ ’)

100*(‘ βirr_dd89’ + 2*`m_irr_dd89’*’β irr_dd89_sq’) Standard errors …. …. …. …. …. .… …. …. 2.Growing season precipitation (inches) (‘ β_dry_prcp’+ 2*`m_dry_prcp’*β_dry_prcp_sq’) (‘ βirr_prcp’+ 2*`m_irr_prcp’*’βirr_prcp_sq’) Standard errors …. …. …. …. …. .… …. …. 14

(16)

4. Results and discussion 4.1 Results

This section presents the output of the four sets of regressions. Firstly, each regression set will be covered separately. Firstly, the coefficients of each regression set are displayed in tables 3a, 4a, 5a and 6a. Consequently, these estimates are used to calculate the marginal estimates in tables 3b, 4b, 5b and 6b. Table 3b uses the coefficients from table 3a in order to calculate the marginal estimates, 4b uses the coefficients in 4a, 5b uses the coefficients in table 5a and finally, 6b uses the coefficients in 6a to calculate the marginal estimates. In section 4.2 a comparison of all the results will be made with respect to the results in table 1 of DG (2012, p.3766). Lastly, section 4.3 will provide a

discussion of the shortcomings of this paper.

4.1.1 Regression 1

Table 3a displays the coefficients for regression 1. This is the regression that uses the four weather variables, nine soil variables, year times regional effects, and elevation as independent variables. To reiterate, the regression coefficients from table 3a are used in the linear equations listed above to obtain the estimates in table 3b.

When taking a closer look at the standard errors in table 3a one can infer that the statistical significance of the temperature variables is very low while the significance of the precipitation variables is much better. Nearly all weather variables suggest that the variation in weather has a negligible effect on profits per acre both in irrigated and non-irrigated counties. Furthermore, the type of yearly times regional effect does not seem to make a significant difference on the estimates. The only weather variables that are consistently significant over all regional times yearly effects are ‘irr_prcp’ an ‘irr_prcp_sq’ which are the weather variables that define precipitation in irrigated counties. These variables are consistently significant at the 1% level with county-clustered standard errors.

(17)

Additionally, the newly added elevation variable has coefficients that range from -0.005 to -0.014 which states that 1m of elevation leads to a decrease in profits per acre by 0.5 to 1.4 dollar cents. These estimates are significant in columns A, C and D but only with state-clustered standard errors.

When looking at table 3b, one can conclude that a temperature increase of a 100 growing season degree days will lead to an increase in profits by $0.19 - $0.85 per acre in non-irrigated counties. In irrigated counties, the increase in growing season degree days will lead to - $0.10 to $1.06 change in profits per acre depending on which yearly times regional effect is included. From row 1 we can infer that an increase in temperature alone, will lead to a very small change – either positive or negative – in profits per acre. Additionally, the statistical significance of these estimates is very low.

As shown in row 2 of table 3b, the effect of an increase in growing season precipitation differs greatly between non-irrigated and irrigated counties. The effect in non-irrigated counties is slightly positive as an increase in growing season precipitation by one inch leads to an increase in profits per acre that ranges from $0.62 to $1.01. The statistical value for the precipitation estimates in non-irrigated counties is however very low as only regressions B and C are significant at the 10% level 3b (in table3b). Conversely, in irrigated counties a marginal increase in precipitation leads to a change in profits by -$4.90 to -$6.55 per acre. These estimates are all significant at the 1% level with county-clustered standard errors.

(18)

Table 3a. Regression coefficients of equation with control for elevation The dependent variable is:

Y Y Y Y

The included yearly*regional effect is:

Yearly (A) USDA x yearly (B) Census x yearly (C) State x yearly (D)

Elevation -0.014 -0.005 -0.010 -0.010 0.009 0.010 0.009 0.009 [0.005]*** [0.005] [0.005]** [0.005]* Irr_dd89 -0.005 0.001 0.007 0.009 0.030 0.031 0.031 0.032 [0.020] [0.018] [0.020] [0.014] M_irr_dd89 756.268 756.268 756.268 756.268 Standard deviation 1616.016 1616.016 1616.016 1616.016 - - - -

Irr_dd89_sq 2.37E-6 1.92E-6 1.47E-6 1.14E-6

3.80E-6 2.96E-6 3.88E-6 3.87E-6

[2.52E-6] [2.41E-6] [2.63E-6] [2.51E-6]

Dry_dd89 0.008 0.016 0.022 0.021 0.008 0.010 0.010** 0.012* [0.010] [0.012] [0.013]* [0.014] M_dry_dd89 3064.761 3064.761 3064.761 3064.761 Standard deviation 1805.647 1805.647 1805.647 1805.647 - - - -

Dry_dd89_sq -1.03E-6 -1.92E-6 -2.28E-6 1.01E-6

6.87E-7 9.49E-7** 8.78E-7*** 0.012**

[1.21E-6] [1.45E-6] 1.49E-6 [1.62E-6]

Dry_prcp 1.29 2.129 1.941 1.06 1.075 1.154* 1.174* 1.38 [1.220] [1.376] [1.302] [1.563] M_dry_prcp 13.314 13.314 13.314 13.314 9.229 9.229 9.229 9.229 - - - - Dry_prcp_sq -0.025 -0.042 -0.035 -2.11E-6 0.023 0.024* 0.025 1.01E-6** [0.028] [0.032] [0.030] [0.035] Irr_prcp -7.202 -5.532 -6.080 -7.456 2.000*** 2.083*** 2.152*** 2.393*** [3.937]* [3.757] [0.030] [4.605] M_irr_prcp 2.497 2.497 2.497 2.497 Standard Deviation 6.496 6.496 6.496 6.496 - - - - Irr_prcp_sq 0.163 0.127 0.141 0.182 0.046*** 0.047*** 0.049*** 0.055*** [0.083]* [0.079] [0.088] [0.099]* R2 _{– value} _0.43 _0.47 _0.45 _0.47 Number of observations 9364 9364 9364 9364 *** significant at 1%, ** at 5%, * at 10%.

Bold indicates regression coefficient

Non-bold indicates county clustered standard errors

Between square brackets […] indicated state clustered standard errors

(19)

Table 3b. – In sample estimates of the effect of growing season weather on farm profits

per acre controlled for state fixed effects and elevation

Non-irrigated Counties (N=7743) Irrigated Counties (N=1621) Marginal effects

1. Growing season degree days x100 0.19 0.40 0.85 0.85 -0.10 0.37 0.88 1.06 Standard error clustered by county 0.49 0.55 0.60 0.66 2.45 2.49 2.55 2.60 Standard error clustered by state [0.40] [0.39] [0.49]* [0.51] [1.64] [1.42] [1.68] [1.58] 2.Growing season precipitation (inches) 0.62 1.01 1.00 0.79 -6.39 -4.90 -5.38 -6.55 Standard error clustered by county 0.50 0.54* 0.55* 0.64 1.78*** 1.87*** 1.92*** 2.13*** Standard error clustered by state [0.54] [0.60]* [0.57]* [0.69] [3.53]* [3.37] [3.71] [4.11] *** significant at 1%, ** 5%, *10% All values in $2002 per acre

Where an increase in 100 growing season degree days leads to a change in profits per acre of 0.19 in row 1 column 1a. In row 2 column 1a, an increase in precipitation by 1 inch will lead to an increase in profits per acre of 0.62

(20)

4.1.2. Regression 2

Secondly, table 4a presents the results from regression 2 that includes elevation as well as the standard deviation of temperature and precipitation as extra geographic variables. The coefficients display similar statistical significance as table 3a. All weather variables with the exception of ‘irr_prcp’ and ‘irr_prcp_sq’ have a negligible effect on the profits per acre and low statistical significance. The included yearly times regional effect does not seem to make a significant difference.

Looking at the extra geographical variables, elevation has coefficients which range from -0.07 to -0.018 which are significant in columns A, C and D with state-clustered standard errors. From this we can conclude that as elevation increases by 1 meter, profits per acre will decrease by about 0.7 to 1.8 dollar cents - which is similarly to table 2. The coefficients of the standard deviation of precipitation range from 0.333 to 0.748 which means that the variability in rainfall from one year to the next has generally led to a slight increase in profits per acre. For temperature, the opposite effect is observed as the variability in temperature from one year to next has led to a very small decrease in profits per acre. Nevertheless, all the coefficients for the variability in weather have low statistical value. Only the variability of precipitation in column D has some statistical relevance at the 10% level.

As far as the estimates in table 4b are concerned, an increase in temperature by 100 growing season degree days leads to a change in profits per acre of -$0.22 to $0.81 in non-irrigated counties and from -$0.68 to $0.85 in irrigated counties. None of these estimates have any statistical significance though. The estimates in row 2 state that an inch of extra rainfall will change profits per acre by $0.26 to $0.78 in non-irrigated land and a change of- $6.51 to -$8.32 per acre in irrigated land. Again, only the estimates for irrigated farmland are statistically significant (at the 1% level). As table 4b depicts, none of the estimates for precipitation in non-irrigated counties are significant.

(21)

Table 4a. Relevant regression coefficients with control for elevation, SD precipitation and SD temperature The dependent variable is:

Y Y Y Y

Elevation -0.018 -0.007 -0.013 -0.012 0.010* 0.011 0.010 0.010 [0.006]*** [0.007] [0.006]** [0.006]* SD precipitation 0.333 0.432 0.468 0.748 0.400 0.372 0.406 0.421* [0.815] [0.680] [0.815] [0.889] SD temperature -0.002 -0.007 -0.005 0.013 0.006 0.007 0.008 0.010 [0.012] [0.008] [0.010] [0.015] Irr_dd89 -0.010 0.001 0.007 0.007 0.035 0.036 0.037 0.037 [0.022] [0.020] [0.023] [0.022] M_irr_dd89 756.268 756.268 756.268 756.268 Standard deviation 1616.016 1616.016 1616.016 1616.016 - - - -

Irr_dd89_sq 2,32E-6 1.38E-6 9.45E-7 1.06E-6

4.52E-6 4.65E-6 4.62E-7 4.65E-6

[2.81E-6] [2.68E-6] [2.96E-6] [2.87E-6]

Dry_dd89 -0.000 0.016 0.022 0.021 0.010 0.012 0.013* 0.014 [0.011] [0.014] [0.014] [0.015] M_dry_dd89 3064.761 3064.761 3064.761 3064.761 Standard deviation 1805.647 1805.647 1805.647 1805.647 - - - -

Dry_dd89_sq -2.74E-7 -2.00E-6 -2.34E-6 -2.12E-6

8.55E-7 1.17E-6* 1.10E-6** 1.22E-6*

[1.34e-6] [1.54E-6] [1.61E-6] [1.81E-6]

Dry_prcp 0.216 1.413 1.281 0.964 1.229 1.327 1.327 1.503 [1.380] [1.434] [1.478] [1.726] M_dry_prcp 13.314 13.314 13.314 13.314 9.229 9.229 9.229 9.229 - - - - Dry_prcp_sq 0.002 -0.021 -0.015 -0.007 0.027 0.028 0.028 0.032 [0.032] [0.032] [0.034] [0.040] Irr_prcp -9.436 -7.421 -7.95 -8.550 2.400*** 2.551*** 2.558*** 2.755*** [4.862]* [4.551] [4.931] [5.367] M_irr_prcp 2.497 2.497 2.497 2.497 Standard Deviation 6.496 6.496 6.496 6.496 - - - - Irr_prcp_sq 0.223 0.183 0.197 0.209 0.058*** 0.060*** 0.060*** 0.064*** [0.104]** [0.097]* [0.105]* [0.114]* R2 - value 0.43 0.46 0.44 0.46 Number of observations 7023 7023 7023 7023 *** significant at 1%, ** at 5%, * at 10%.

(22)

Table 4b. – In sample estimates of the effect of growing season weather on farm profits

controlled for state fixed effects, elevation and the standard deviation of temperature and

precipitation

Non-irrigated Counties (N=7743) Irrigated Counties (N=1621) Panel A. Marginal

effects

1. Growing season degree days x100 -0.22 0.35 0.82 0.81 -0.68 0.32 0.84 0.85 Standard error clustered by county 0.63 0.68 0.73 0.78 2.89 2.91 3.00 3.05 Standard error clustered by state [0.43] [0.46] [0.53] [0.55] [1.81] [1.62] [1.86] [1.76] 2. Growing season precipitation (inches) 0.26 0.86 0.88 0.78 -8.32 -6.51 -6.97 -7.51 Standard error clustered by county 0.55 0.61 0.60 0.69 2.13*** 2.27*** 2.28*** 2.45*** Standard error clustered by state [0.60] [0.63] [0.88] [0.73] [4.35]* [4.07] [4.41] [4.80] *** significant at 1%, ** 5%, *10% All values in $2002 per acre

Where an increase in growing season degree days leads to a decrease in profits of $0.22 in row 1, column 1a. And an increase in growing season precipitation by 1inch will lead to an increase of profits of $0.26 in row2, column 1a.

(23)

4.1.3. Regression 3

Regression 3 uses all weather, year times regional and soil variables as well as elevation and elevation times temperature as independent variables. See table 5a for the regression coefficients. Like regression 1 and 2, all the temperature variables have low statistical value. The precipitation variables in non-irrigated counties provide estimates for regressions B and C which are significant at the 5% and 10% level respectively. Additionally, the precipitation variables in irrigated counties are again all significant at the 1% level. This would imply that only changes in precipitation in irrigated counties significantly affect profits per acre.

Looking at the extra geographic variables, elevation is significant with county-clustered standard errors in columns A and B and somewhat significant in columns C and D. The coefficients of the elevation variable range between -0.029 and -0.041 implying that an increase in elevation by 1 meter reduces profits per acre by about 3 to 4 dollar cents. Furthermore, the elevation times temperature variable is only significant in columns A and B. Also the coefficients suggest that the effect of elevation with constant temperature is negligibly small.

Finally, looking the estimates in table 5b an increase in growing season degree days by 100 will change profits per acre by -$0.75 to $0.13 in non-irrigated counties and by -$2.13 to -$0.48 in irrigated counties. However, these estimates have poor statistical significance.

As far as a change in 1 inch of growing season precipitation is concerned, in non-irrigated counties profits increase by $0.88 to $1.24 per acre and in irrigated counties profits will change by $4.72 to -$6.40 per acre. The estimates for the non-irrigated counties are only significant at the 5% level in regressions B and C but the ones for irrigated counties are all significant at the 1% level (see row 2 in table 4b).

(24)

Table 5a. Relevant regression coefficients with control for elevation and elevation*temperature The dependent variable is:

Y Y Y Y

Elevation -0.041 -0.034 -0.029 -0.029

0.014*** 0.015** 0.015* 0.016*

[0.020]** [0.021] [0.022] [0.024]

Elevation*temp 7.94E-6 7.86E-6 5.23E-6 5.42E-6

4.44E-6* 4.43E-6* 4.71E-6 5.06E-6

[5.15E-6] [5.27E-6] [5.51E-6] [6.10E-6]

Irr_dd89 -0.028 -0.022 -0.010 -0.009 0.031 0.032 0.033 0.033 [0.017] [0.019] [0.018] [0.020] M_irr_dd89 756.268 756.268 756.268 756.268 Standard deviation 1616.016 1616.016 1616.016 1616.016 - - - -

Irr_dd89_sq 4.21E-6 3.71E-6 2.78E-6 2.55E-6

3.90E-6 4.03E-6 3.97E-6 3.97E-6

[2.03E-6]** [2.08E-6]* [2.13E-6] [2.21E-6]

Dry_dd89 -0.012 -0.006 0.007 0.005 0.012 0.015 0.015 0.018 [0.017] [0.019] [0.020] [0.023] M_dry_dd89 3064.761 3064.761 3064.761 3064.761 Standard deviation 1805.647 1805.647 1805.647 1805.647 - - - -

Dry_dd89_sq 6.77E-7 -1.29E-7 -9.75E-7 -6.41E-7

1.10E-6 1.35E-6 1.35E-6 1.57E-6

[6.77E-7] [1.73E-6] [1.94E-6] [2.25E-6]

Dry_prcp 1.757 2.536 2.195 1.332 1.176 1.213** 1.257* 1.470 [1.311] [1.407]* [1.335] [1.623] M_dry_prcp 13.314 13.314 13.314 13.314 9.229 9.229 9.229 9.229 - - - - Dry_prcp_sq -0.033 -0.049 -0.039 -0.014 0.024 0.025* 0.026 0.030 [0.029] [0.032] [0.031] [0.036] Irr_prcp -6.880 -5.35 -5.938 -7.32 2.051*** 2.095** 2.181*** 2.420*** [3.973]* [3.773] [4.169] [4.639] M_irr_prcp 2.497 2.497 2.497 2.497 Standard Deviation 6.496 6.496 6.496 6.496 - - - - Irr_prcp_sq 0.162 0.128 0.141 0.183 0.046*** 0.047*** 0.049*** 0.055*** [0.084]** [0.080] [0.090] [0.101]* R2 _{- value} _0.44 _0.47 _0.45 _0.47 Number of observations 9364 9364 9364 9364 *** significant at 1%, ** at 5%, * at 10%.

(25)

Table 5b. – In sample estimates of the effect of growing season weather on farm profits

**controlled for state fixed effects, elevation and elevation*temperature (in $/ acre)**

effects

Growing season degree days x100 -0.75 -0.65 0.13 0.08 -2.13 -1.68 -0.57 -0.48 Standard error clustered by county 0.60 0.75 0.77 0.86 2.55 2.63 2.70 2.77 Standard error clustered by state [0.77] [0.92] [0.92] [0.94] [1.48] [1.58] [1.55] [0.78] Growing season precipitation (inches) 0.88 1.24 1.15 0.95 -6.07 -4.72 -5.23 -6.40 Standard error clustered by county 0.56 0.58** 0.60** 0.70 1.84*** 1.88*** 1.95*** 2.17*** Standard error clustered by state [0.61] [0.63]* [0.60]** [0.74] [3.56]* [3.38] [3.73] [4.14] *** significant at 1%, ** 5%, *10% All values in $2002 per acre

Where an increase in growing season precipitation by 100 lead to a decrease in profits per acre by $0.75. And an increase in growing season precipitation by 1 inch will lead to an increase in profits by $0.88 per acre.

(26)

4.1.4. Regression 4

Finally, regression 4 incorporates all the variables in one analysis. In table 6a, the temperature variables again do not portray consistent statistical significance whereas the precipitation variables are only significant in irrigated counties. For temperature, only ‘irr_dd89’ and ‘dry_dd89’ are

somewhat significant in column A. On the other hand, precipitation in irrigated counties is significant over all year times regional effects at the 1% level. This is a similar trend to tables 2-4.

In the category of extra geographic variables only elevation consists of significant coefficients which range from -0.035 to -0.052. Elevation times temperature has coefficients that are significant at 10% in columns A and B which are respectively: 9.79E-6 and 8.89E-6. We can thus infer again that the interaction between elevation and temperature only has a minor effect on profits. The coefficients for the variability of the weather variables are very similar to table 3 and also have low statistical value.

Table 6b presents the estimates that have been deduced from table 5a. The estimated marginal effect of a temperature increase of a 100 growing season degree days range from -$1.37 to $0.07 per acre in non-irrigated counties. In irrigated land, these marginal effects range from -$3.08 to -$0.63 per acre. In row 2, with growing season precipitation, an increase of rainfall by 1 inch will lead to an increase of profits by $0.65 to $1.12. For irrigated land, the marginal effects range between $-7.86 to - $6.31 per acre. As is indicated in table 5b, only the estimates for precipitation in farmland are significant.

(27)

Table 6a. Regression coefficients with control for elevation, SD temperature, SD precipitation and elevation*temperature

The dependent variable is:

Y Y Y Y

Elevation -0.052 -0.039 -0.035 -0.032

0.016*** 0.018** 0.018* 0.019

[0.023]** [0.024] [0.025] [0.027]

Elevation*temp 9.79E-6 8.89E-6 6.06E-6 5.45E-6

5.17E-6* 5.01E-6* 5.42E-6 5.69E-6

[5.95E-6] [6.08E-6] [6.39E-6] [6.87E-6]

SD temperature -0.007 -0.012 -0.008 0.009 0.006 0.008 0.008 0.010 [0.010] [0.006]* [0.008] [0.013] SD precipitation 0.322 0.443 0.477 0.763 0.397 0.370 0.405 0.420* [0.797] [0.665] [0.807] [0.878] Irr_dd89 -0.038 -0.024 -0.011 -0.010 0.037 0.037 0.038 0.039 [0.019]** [0.022] [0.020] [0.022] M_irr_dd89 756.268 756.268 756.268 756.268 Standard deviation 1616.016 1616.016 1616.016 1616.016 - - - -

Irr_dd89_sq 4.46E-6 3.31E-6 2.37E-6 2.38E-6

4.65E-6 4.74E-6 4.72E-6 4.78E-6

[2.12E-6]** [2.29E-6] [2.37E-6] [2.49E-6]

Dry_dd89 -0.025 -0.008 0.006 0.005 0.015* 0.018 0.018 0.020 [0.019] [0.023] [0.023] [0.027] M_dry_dd89 3064.761 3064.761 3064.761 3064.761 Standard deviation 1805.647 1805.647 1805.647 1805.647 - - - -

Dry_dd89_sq 1.79E-6 -4.19E-8 -8.88E-7 -7.43E-7

1.31E-6 1.59E-6 1.58E-6 1.79E-6

[1.87E-6] [2.05E-6} [2.20E-6] [2.62E-6]

Dry_prcp 0.905 1.881 1.604 1.257 1.379 1.391 1.434 1.611 [1.528] [1.483] [1.552] [1.805] M_dry_prcp 13.314 13.314 13.314 13.314 9.229 9.229 9.229 9.229 - - - - Dry_prcp_sq -0.009 -0.029 -0.020 -0.011 0.029 0.030 0.030 0.033 [0.034] [0.033] [0.035] [0.041] Irr_prcp -8.961 -7.227 -7.782 -8.40 2.478*** 2.557*** 2.591*** 2.785*** [4.913]* [4.574] [4.960] [5.399] M_irr_prcp 2.497 2.497 2.497 2.497 Standard Deviation 6.496 6.496 6.496 6.496 - - - - Irr_prcp_sq 0.220 0.184 0.197 0.209 0.058*** 0.059*** 0.060*** 0.064*** [0.106]** [0.099]* [0.107] [0.117]* R2 - value 0.43 0.46 0.44 0.46 Number of observations 7023 7023 7023 7023 *** significant at 1%, ** at 5%, * at 10%.

(28)

Table 6b. – In sample estimates of the effect of growing season weather on farm profits

**controlled for: state fixed effects, elevation, elevation*temperature and the standard**

deviation of temperature and precipitation

effects

1. Growing season degree days x100 -1.37 -0.81 0.02 0.07 -3.08 -1.93 -0.76 -0.63 Standard error clustered by county 0.76* 0.89 0.91 1.00 3.01 3.06 3.16 3.26 Standard error clustered by state [0.86] [1.06] [1.01] [1.14] [1.58] [1.85] [1.73] [1.88] 2. Growing season precipitation (inches) 0.65 1.12 1.06 0.95 -7.86 -6.31 -6.80 -7.35 Standard error clustered by county 0.64 0.66* 0.67 0.77 2.21*** 2.28*** 2.31*** 2.49*** Standard error clustered by state [0.70] [0.67] [0.69] [0.79] [4.39] [4.08] [4.43] [4.82] *** significant at 1%, ** 5%, *10% All values in $2002 per acre

Where an increase in 100 growing season degree days in row 1, column 1a leads to a change in profits by -$1.37. And an increase in growing season precipitation by 1 inch leads to a change of $0.65 in row 2, column 1a.

(29)

4.2 Comparison with DG (2012)

Table 7 depicts table 1 of DG (2012, p.3766) with county fixed effects and no extra geographic variables. In row 1 the differences are small. In non-irrigated counties the marginal effect of an increase of 100 growing season degree days varies between -$1.27 and $0.10. For irrigated counties the effects lie between -$1.51 and $1.48. These ranges do not differ greatly to tables 3b, 4b, 5b and 6b. Interestingly, in column A of non-irrigated counties, the marginal effect an increase of 100 growing season degree days appears to be significant at the 1% level. Thus, one could argue that estimating the effects of an increase in temperature is more efficient with county fixed effects and no extra geographic variables. Conversely, the estimates for regressions B, C and D provide no significant estimates suggesting that this trend only holds when yearly fixed effects are included. Subsequently, for irrigated counties none of the estimates are statistically significant.

Table 7. DG (2012) results – In sample estimates of the effect of growing season weather

on form profits based on corrected data

effects

1. Growing season degree days x100 -1.27 -0.39 0.10 -0.18 -1.51 0.87 1.48 1.09 Standard error clustered by county 0.25*** 0.35 0.37 0.75 2.19 1.87 2.17 2.87 Standard error clustered by state [0.39]*** [0.41] [0.54] [0.56] [1.81] [1.60] [1.82] [2.48] 2. Growing season precipitation (inches) -0.58 0.61 0.13 0.03 0.06 2.30 1.77 0.91 Standard error clustered by county 0.24** 0.24** 0.23 0.26 1.62 1.59 1.56 1.53 Standard error clustered by state [0.55] [0.39] [0.35] [0.33] [1.48] [1.27]* [1.45] [1.69] *** significant at 1%, ** 5%, *10% All values in $2002 per acre

Where an increase in 100 growing season degree days in row 1, column 1a leads to a change in profits by -$1.37. And an increase in growing season precipitation by 1 inch leads to a change of $0.65 in row 2, column 1a.

(30)

When looking at the marginal effects of growing season precipitation in row 2, for non-irrigated counties the estimates seem to be slightly more negative compared with tables 3b, 4b, 5b and 6b. Nevertheless, the differences are small. Again, in table 7 the statistical significance for the regression with yearly effects are higher in column 1A than in the previous tables. A potential cause of this trend is the fact that year times regional effects may cause collinearity in combination with county fixed effects leading to inflated standard errors for regressions B, C and D. Yet, the fact that this trend is only observed in non-irrigated counties weakens this theory. The most striking difference between table 7 and the previous tables is observed for an increase in growing season precipitation in irrigated counties. Where table 7 provides moderately positive estimates with low statistical significance, tables 3b-6b display significantly negative estimates. Essentially, we can conclude that the inclusion of state fixed instead of county fixed effects, and control for elevation, weather variability and the interaction between temperature and elevation in the model, significantly worsens the marginal effect of an increase in growing season precipitation .

4.3 Discussion

This paper provides a more thorough insight on the effect of climate change on the agricultural sector by including elevation, weather variability and an interaction variable for elevation and temperature. Nevertheless, some problematic issues remain. These will be covered below.

Firstly, the current data and model specification still suffer from collinearity issues. The underlying cause of this issue is the fixed effect regression technique that is applied. Where DG (2007), Fisher et al. (2012) and DG (2012) all absorb for county fixed effects in order to compensate for prices and output differences, they also appear to be absorbing weather and elevation variation. In order to reduce the extent of the collinearity problem, this paper uses state fixed effects instead. This adjustment seems to improve the statistical value of the estimates yet as more estimates are significant, yet some year times regional fixed effects are still omitted by STATA due to collinearity.

(31)

More collinearity issues arise between the variables as the variables between irrigated and non-irrigated counties show hints of linear correlation. So, while the distinction between non-non-irrigated and irrigated land is relevant for analyzing the agricultural sector, it also presents some statistical difficulties. Furthermore, in theory, temperature, precipitation and elevation are also not completely independent. Elevation is known to stimulate precipitation and effect temperature. While

temperature in turn is known to effect precipitation levels (Beniston et al., 1997, p. 333-235).

Moreover, the collinearity issues and the adjustment for state fixed effects also leads to difficulties in analyzing the roles the extra geographic variables have on profits. In effect, with the current

regression techniques, model and data, it is difficult to accurately determine the marginal effect of each independent variable. Further research on how to avoid these collinearity issues is required in order to improve the estimates of this model.

Lastly, the USA is arguably not the best case study to research the link between elevation and profits per acre as the USA only consists of one significant mountain range – the Rocky Mountains in

Colorado (see figure 1.). Another region with more variation in elevation could yield better results. Also, the current data is at county level for nearly the entire USA. Adjusting for all the county or even state fixed effects leads to the collinearity issues described above. Similar research in a less

aggregated area could provide different insights on the marginal effects of climate change on agricultural profits.

(32)

Figure 1. Average elevation per county in mainland USA (elevation in meters) USA elevation data

(In meters)

(33)

5. Conclusion

To conclude, previous research by DG (2007), Fisher et al. (2012) and DG (2012) provide a thorough basis for estimating the effects of climate change on the agricultural sector. The data and appendices they provide facilitate further research on this topic. Moreover, Beniston et al. (1997) and Frauenfeld et al. (2005) emphasize the important impact mountains and elevated regions have on climate change. Beniston et al. (1997) furthermore stress that sudden changes in weather variability are likely to have more impact than mere changes in mean weather levels. In order to improve the estimates by DG (2012), this paper adds new geographic variables to the DG data appendix provided. In doing so, new estimates – controlled for differences in elevation and changes in weather

variability – are provided.

This paper provides marginal estimates for changes in temperature and precipitation based on four regression analyses – each controlled for different extra geographic variables. These estimates are provided in tables 3b, 4b, 5b, and 6b. When looking at the growing season degree days in row 1, the estimates become increasingly negative from table 3b to 6b. In other words, control for weather variability and the interaction variable elevation times temperature appears to have worsen the marginal effects of an increase in 100 growing season degree days. The standard errors of the estimates are not statistically significant though, suggesting that we cannot take this trend for granted.

Taking a closer look at yearly times regional fixed effect in each column, columns A and B appear to have much lower estimates than columns C and D. In effect, including yearly and year times USDA regional effects seems to have a more negative effect on y than year times census division or year times state fixed effects seem to have.

Ultimately, considering the fact that $65.56 per acre is the mean of y, an increase of a 100 growing season degree days will only have a minor effect on profits per acre. In row 1, the largest positive

(34)

marginal estimate observed is $0.85 for both non-irrigated and irrigated counties (in tables 2b and 3b respectively) whereas the largest negative marginal estimate for non-irrigated land is -$1.37 and -$3.08 in irrigated land (see table 5b). With respect to the mean of y, this accounts for only a 1.3% increase in profits in the most fortunate of outcomes or a 2,1% to 4.7% decrease in the most disastrous of outcomes.

When looking at precipitation in row 2, it is interesting to observe the contrast in marginal estimates between non-irrigated and irrigated counties. While an increase of an inch in precipitation leads to a minor increase in profits per acre in non-irrigated counties, the opposite effect can be observed in irrigated counties where the marginal effects are significantly negative. However, it is crucial to note that the standard errors for non-irrigated counties are less significant than the estimates for irrigated land.

As far as the year times regional effects are concerned, columns A and D with yearly and year times state fixed effects, appear to have the lowest estimates whereas the columns B and C provide more positive approximations.

In the most fortunate of outcomes, an increase of 1 inch in growing season precipitation leads to a $1.24 per acre increase in non-irrigated land and $4.72 decrease in profits per acre in irrigated land (see table 4b). In percentages, this amounts to a 1.9% increase or a 7.2% decrease of profits with respect to the mean of y. In the most disastrous of outcomes, an increase of $0.26 or a decrease of $8.32 per acre will occur in non-irrigated and irrigated counties respectively. Correspondingly, this amounts to a 0.4% increase and a 12.7% decrease.

To conclude, in irrigated regions we can assume that i) an increase in growing season precipitation will lead to substantial losses in the agricultural sector and ii) that an increase in temperature will either lead to minor profits or losses. In non-irrigated counties, we expect that i) an increase in

(35)

growing season precipitation will result in a small increase in profits. Whilst ii) an increase in temperature can result in either a minor increase in profits or minor losses.

Overall, the inclusion of elevation and variability in this model does not produce unrealistic estimates. An increase in precipitation in irrigated regions could lead to an excess of water and flooding of farmland. Conversely, in dry and non-irrigated regions extra rainfall would provide crops with a much needed water supply. Unfortunately, we cannot deduce much from the ambiguousness of the temperature estimates aside for the fact that the marginal effect on profits are likely to be small. Much still remains to be explored in the field of climate change and thus more research and new data is required in order to overcome the collinearity issue described in the discussion.

(36)

References

- Beniston, M., Diaz, H. F., & Bradley, R. S. (1997). Climatic change at high elevation sites: an overview. Climatic Change 36 (3-4): 233-251.

- Deschênes, O., & Greenstone, M. (2007). “The economic Impacts of climate change: evidence from Agricultural Output and Random Fluctuations in Weather." The American Economic Review 97 (1): 354–85.

- Deschênes, O., & Greenstone, M. (2012). “The Economic Impacts of Climate Change: Evidence from Agricultural Output and Random Fluctuations in Weather: Reply.” The American Economic Review 102 (7): 3761-3773.

- EROS. (2010). “30 arc-second DEM of North America (Data Basin Dataset).” Retrieved June 10, 2013 from:

http://www.arcgis.com/home/item.html?id=5771199a57cc4c29ad9791022acd7f74 - Fisher, A. C., Hanemann, W. M., Roberts, M. J., & Schlenker, W. (2012). “The economic

impacts of climate change: evidence from agricultural output and random fluctuations in weather: comment.” The American Economic Review 102 (7): 3749-3760.

- Frauenfeld, O. W., Zhang, T., & Serreze, M. C. (2005). “Climate change and variability using European Centre for Medium-Range Weather Forecasts reanalysis (ERA-40) temperatures on the Tibetan Plateau.” Journal of Geophysical Research 110 (D2): D02101.

- McKibbin, W. J., & Wilcoxen, P. J. (2002). “The role of economics in climate change policy.” Journal of economic perspectives 16 (2): 107-129.

- You, Q., Kang, S., Pepin, N., & Yan, Y. (2008). “Relationship between trends in temperature extremes and elevation in the eastern and central Tibetan Plateau, 1961–2005.” Geophysical Research Letters 35(4).