Economic evaluation of climate chang impacts on road transportation in Atlantic Canada

Climate change impacts such as an increase in mean temperature, change in precipitation patterns and sea level rise are affecting regional road transportation network (RRTN) in Atlantic Canada. Those impacts cause direct and indirect economic consequences for the network and regional economy. In our study, we constructed a dynamic computable general equilibrium model (CGEM) to trace these consequences over time. Basic principles of the designed CGEM are discussed and the model's architecture is presented. The model's elements are estimated, and the obtained CGEM is tested with exogenously imposed shock. Evolutionary dynamics of regional temperature, precipitation and sea level is analyzed on the basis of comprehensive time series analysis. This dynamics will be later imposed on the designed CGEM as external productivity shocks. Some preliminary cumulative economic consequences are evaluated in monetary terms to obtain benchmarks for the mitigation measures associated with future development of the RRTN.


INTRODUCTION
The Fifth Assessment Report (AR5) of the Intergovernmental Panel on Climate Change emphasized that previous assessments have already shown through multiple lines of evidence that the climate is changing across our planet, largely as a result of human activities [1] [2]. The most compelling evidence of climate change derives from observations of the atmosphere, land, oceans and cryosphere.
According to AR5, the main climatic drivers in North America are temperature warming and drying trends, extreme temperature events, extreme precipitation events, damaging cyclones and rising sea level. Therefore, in this study three fundamental climate change impacts namely increasing temperature, changing precipitation patterns and rising sea level have been analyzed with respect to their potential economic consequences for the Regional Road Transportation Network (RRTN) in Atlantic Canada.
Our RRTN for this study was defined according to guidelines of the New Brunswick Department of Transportation presented in "Charting the Course: Atlantic Canada Transportation Strategy, 2008-2018", and provincial Economic Development and Innovations Plan. Five regional transportation hubs -Fredericton, Moncton, St. John, Northern New Brunswick (Edmundston) and Halifax -were chosen on the basis of the above mentioned documents.
Our study was dedicated to the following three tasks: Design of the model's architecture to incorporate microeconomic factors, regional macroeconomic factors and climate change impacts on the basis of systems dynamics Estimation of the model's elements on the basis of dynamic analysis Analysis of the evolutionary dynamics of climate change variables -temperature, precipitation and sea level -associated with five regional hubs In order to perform all these tasks, comprehensive economic and transportation database has been developed. Below the results of our study are presented according to the above mentioned tasks.

II. MODEL'S ARCHITECTURE
Our analysis of the existing literature on economic evaluation of climate change impacts in various segments of an economic system, led us to the conclusion that dynamic general equilibrium model (GEM) is the best framework for our analysis. Reference [3] is a seminal work in this regard followed by more recent studies presented in [4]- [6]. These studies helped us formulate basic principles for our model: Both microeconomic and macroeconomic aspects of the RRTN should be incorporated simultaneously Modified Computable General Equilibrium Model (CGEM) is our fundamental tool since it allows us to directly incorporate micro-and macroeconomic dynamics as well as dynamics of climate change impacts Modeling has to address the following two goals: (i) design of the appropriate model architecture to include microeconomic and macroeconomic dynamics affecting our RRTN; (ii) modeling of the dynamics of climate change variables.
Once the above goals are achieved, dynamics of the climate change variables should be imposed on the dynamics of the basic CGEM to trace economic consequences of the climate change impacts. Below our model's architecture is described in detail.
Regional road transportation module is the centerpiece of our CGEM, and it represents the highest level 1 in our architecture. It consists of a system of dynamic equations for price and volume of regional transportation (traffic).Within the module, this system produces equilibrium price of regional transportation for the entire RRTN based on major economic determinants explained in detail in the next section.
Next level 2 consists of five regional hubs defined previously -Fredericton, Moncton, St. John, Edmundston and Halifax. At each hub, the above price of transportation is taken as an input to define the value added produced by the hub based on this price, geographical and industrial characteristics as well as the required volume of transportation (traffic).
Each hub is further disaggregated with respect to major industrial consumers of transportation in that sub-region which defines our level 3 -the lowest one. At this level, the value added produced by each major industry is defined on the basis of microeconomic and regional macroeconomic characteristics and then is sent to level 2.
At level 2, aggregate value added by each industry is summed up at corresponding hub, the hub's traffic to support this aggregate value added is defined and both values are sent back to level 1 to determine next period equilibrium price of transportation for the entire RRTN. This process repeats itself to produce time paths of our major economic and transportation variables associated with the RRTN over time given current and expected micro-and macroeconomic conditions.
In line with the CGEM framework described in the literature, eventually we obtain a time path of the system under study which is a set of short-run equilibria of that system. Microeconomic forces are internal in our model since they are based on the supply-demand dynamics of the regional industries/sectors while macroeconomic forces are based on external dynamics given exogenously. On top of that, dynamics of climate change variables should be imposed as well. In other words, given initial conditions, our model replicates evolution of the RRTN, driven by internal microeconomic forces plus external macroeconomic forces and climate change impacts.
According to the philosophy of the dynamic CGEM, all model's elements should be estimated on the basis of statistical analysis and historical data outside the model. Below we present this procedure in detail beginning with the RRTN dynamics and followed by the climate change dynamics.

III. ESTIMATION AND SIMULATION
A. Main module estimation Regional road transportation module represents a dynamic version of supply-demand system with equilibrium price and traffic level for the entire RRTN as endogenous variables. Our choice of independent or exogenous variables was based on fundamental forces of demand and supply. Regional gross domestic product (GDP) and value added by largest regional consumers of transportation are important demand side determinants while oil price is major supply side determinant.
In addition, population and general price level presented by the consumer price index were added to address the issue of nominal and real values in our demand-supply model.
In dynamic setting, our structural demand-supply system can be presented as follows: where Tt D and Tt S are volume of transportation (traffic) demanded and supplied respectively at time t; Pt T is the price index for regional transportation; CPIt is the consumer price index; GDPt is regional GDP; OILt is oil price, VAt is the value added by the largest regional consumers of transportation; POPt is regional population; et D and et S are demand and supply shocks respectively.
Solved for equilibrium values of Tt * and Pt T* , initial structural demand-supply system produces the following system of reduced form equations:  (2) in which u1t and u2t are composite shocks.
In econometrics, such systems are usually estimated with the help of the Vector Autoregression (VAR). This procedure is described in all details in [7], and we followed this methodology.
Regional transportation data were provided by the New Brunswick Department of Transportation and Nova Scotia Transportation and Infrastructure Renewal. The traffic count for major roads in our RRTN was taken daily and then averaged on an annual basis. Economic data were provided by Transport Canada and Statistics Canada specifically from Canadian economic database CANSIM. Obtained data covers period of 1990-2013.
All our variables were transformed into natural logarithms. It is the usual procedure in econometrics because of the following two reasons. First, logarithmic transformation addresses non-linearity between dependent and independent variables. In such a case, non-linear relationship in levels is replaced by linear relationship in logarithms. Second, economic interpretation of the coefficients becomes more convenient: they reflect a percentage change in dependent variable due to 1% change in independent variable.
Our VAR was estimated in Eviews 8, and bellow we present our results: Adjusted R-squared of the above regression is 0.96, and all coefficients are statistically significant. These results show that dynamics of the demand-supply system was captured well. B. Regional hubs estimation At each regional hub, four largest consumers of transportation were identified: (i) forestry and logging, (ii) wholesale trade, (iii) retail trade, and (iv) manufacturing. Consequently, each hub was disaggregated to include microeconomic dynamics of these industries.
In an economic sense, each industry at a corresponding hub can be presented by dynamic demand-supply system. Similar to the previous explanation, if these demand-supply systems are solved for equilibrium values of price and quantity produced by each industry, in terms of time series analysis we end up with Panel Vector Autoregression (PVAR): where VAi is value added produced by the i-th industry, Pi is the production price index of the i-th industry, Di is industry dummy: 1 for forestry and logging, 2 for wholesale trade, 3 for retail trade, 4 for manufacturing; all other variables were already explained. In this specification it is assumed that at each hub microeconomic dynamics is similar but each industry contributes its own value added based on its own price index, common regional characteristics and overall price of transportation for the whole RRTN.
Since Fredericton and Halifax transportation hubs are the most influential among all five hubs, we have estimated statistical model (4) for those hubs as follows:  for Halifax hub. Adjusted R-squared is 0.80 and 0.81 respectively. Once all elements of our CGEM were evaluated statistically, we used the overall model to simulate climate change impacts on our RRTN.

C. Simulation
In our previous study [8], we evaluated annual impact from climate change on traffic produced in our RRTN. Expected value of the annual loss in traffic due to climate change in Atlantic Canada expressed through increasing temperature, changing precipitation and rising sea level was defined as 1.5%. Therefore, we imposed this negative quantity shock on our CGEM to trace its consequences for the RRTN over time. According to our simulation, immediate impact of a 1.5% loss in regional traffic due to climate change impacts leads to a 1.8% loss in the regional value added or $0.959 billion in dollars of 2007. Over time this impact accumulates and by the tenth year annual loss in regional value added becomes 8.2% or $4.367 billion. Long run cumulative loss is 19.4% or $10.332 billion. In terms of traffic produced in the RRTN, short-run loss in traffic is 2.9% while cumulative longrun loss is 31.8%.
The above values were defined under assumption that no mitigation measures would be taken. On the one hand, these numbers point to significant climate change impact on regional economy but on the other hand they can serve as benchmarks for investment in mitigation measures. Of course, in order to derive marginal impacts from rising temperature, changing precipitation and rising sea level on our RRTN, it is necessary to study dynamics of these climate variables first and then impose it directly on our CGEM. We have already done some preliminary evaluation of dynamics of climate variables which is presented below.

IV. EVOLUTIONARY DYNAMICS OF CLIMATE VARIABLES
As stated before, three fundamental climate variablestemperature, precipitation and sea level -have been chosen for our analysis of climate change dynamics. Our analysis of evolutionary dynamic of these variables is based on stationary linear models widely used in time series econometrics: autoregressive moving-average (ARMA) and generalized autoregressive conditional heteroskedastic (GARCH) models.
We assumed that dynamic path of a climate variable can be described by a dynamic process in the form of a linear stochastic difference equation. Accordingly, our estimation procedure involved the following steps: (i) identification and estimation of the baseline ARMA model; (ii) series of parameter instability tests; (iii) identification and estimation of appropriate error term structure for ARMA model with particular focus on ARCH and GARCH specifications; (iv) selection of the best fitting model followed by in-sample (onestep-ahead) and out-of-sample (dynamic) forecast, and series decomposition. Below we present our major findings

A. Temperature
We used monthly mean of daily temperature time series data obtained from the second generation of adjusted and homogenized Canadian climate dataset (AHCCD). AHCCD was constructed by the Meteorological Service of Canada specifically for the use in climate change research and trend analysis. It is based on historical temperature records from National Climate Data and Information Archive of Environment Canada. One advantage of the AHCCD is that it contains long time series data, which was obtained by combining, homogenizing and adjusting observations from colocated meteorological stations. AHCCD covers 338 locations across Canada including 46 locations in Atlantic Canada. A complete overview of the dataset including statistical adjustment algorithms and data correction methods is presented in [9]. We chose stations located close to our five regional hubs.
Based on the above presented four steps, we concluded that irrespective of the season of the year mean monthly temperature will be increasing on average by +0.0011755°C during 1872-2101. Estimated cumulative increase in mean monthly temperature will amount to +3.25°C by the end of this century if compared with initial temperature level of 1872 or to +1.43°C if compared with 2000. According to our estimates, cumulative increase in mean monthly temperature amounts to +1.95°C between 1872 and 2010. These values correspond to the estimates reported in AR5. All details of this analysis can be found in [10].
So, our statistical analysis of the temperature dynamics supports a stylized fact that climate will be getting warmer in Atlantic Canada by the end of the 21 st century. Most importantly it gives us numerical value of the speed of the increasing temperature.

B. Precipitation
Our precipitation data is taken from the second generation Adjusted Precipitation for Canada (APC2) dataset. Data is annual, obtained from daily values for rainfall in millimeters (mm) and snowfall in centimeters (cm). As previously, we took the data from meteorological stations that are close to our five regional transportation hubs. According to our methodology described at the beginning of this section, we can conclude the following. There are currently stable positive trends in rainfall in Fredericton and Moncton: annual increase in rainfall is 2. Fredericton rainfall time series has higher than average variation after mid-1990s. Moncton and Saint John rainfall series exhibit upward shift in variation in 1950s which stays at higher level since then.

C.
In snowfall time series, variation is more controversial. Miramichi, Moncton, and Halifax snowfall series exhibit a declining variation. Miramichi time series has declining variation since 1950s, Moncton series variation has declined since 1980s, Halifax time series has sharp decline in variation in 1970s with low level afterwards. Edmundston, Fredericton, and Saint John snowfall series have increasing variation. Edmundston series variation has grown since 1980s; Saint John and Fredericton series have sharp increase in variation in 1960s, which stayed at high level afterwards. Detailed analysis of precipitation dynamic is presented in [11]. For most of our regional transportation hubs, rain has higher share in total precipitation. Except Halifax and Moncton, where patterns are not clear, other regional hubs exhibit high and increasing variation in the second part of 20 th century. Increasing variation means less predictable weather coupled with an increase in probability of extreme precipitation events.

D. Sea level rise
In general, analysis of sea level rise is a complex process subject to various factors including vertical land motion, glacier isostatic adjustment, regional oceanographic effects, extreme water levels and others. Since a very complicated geophysical process underlies a sea level rise, its explanation is beyond our analysis. In this study, as with our temperature and precipitation analysis, we have tried to capture dynamics of sea level rise using linear stochastic time series models.
The data provided by Canadian Hydrographic Service (a division of the Science Branch of the Department of Fisheries and Oceans Canada) was chosen as our main source. Monthly mean sea level data for three tide-gauge stations were downloaded from the Atlantic Zone Monitoring Program (AZMP). All stations were chosen according to geographical location of the regional transportations hubs identified in our study as RRTN.
Based on our methodology, we were able to make some forecasts for projected sea levels for 2100: Fredericton hub is not affected by the sea level since it is located too far from the Atlantic Ocean. As a matter of fact, obtained results are consistent with estimates presented in Updated Sea-Level Rise and Flooding Estimates for New Brunswick Coastal Sections Based on AR5.
We also found accelerating volatility in the last decade in sea level data which leads to a conclusion that frequency of extreme events associated with sea level rise is increasing, and we should take it into account in our future forecast.
V. CONCLUSIONS Major result of our study is design of a suitable methodology to trace economic consequences of climate change impacts on regional road transportation network (RRTN). The designed methodology is based on systems approach namely the so-called systems dynamics. In order to implement basic principles of systems dynamics it was necessary to define our system first. We defined it as a hierarchical, three-level model with some specific architecture. Technically it is a modified version of the computable general equilibrium model (CGEM). However, traditional CGEMs are either macroeconomic models with aggregate markets/sectors or microeconomic models when economy is viewed as a collection of markets for goods and services (one good/service -one market) which are simultaneously in equilibrium. Our model is somewhere in between these two types of CGEM since we are modeling RRTN and therefore, microeconomic and macroeconomic components were incorporated into quasispatial model of a transportation network.
All CGEMs are first estimated on the basis of statistical analysis and then are used for simulation purposes. In our study, we have been collecting regional economic and transportation data in order to estimate our model's elements statistically. The designed model's architecture requires a lot of specific datasets which is very time consuming. This is a continuing process which we have been doing since 2012, and we managed to construct our own database. This database allowed us to do some preliminary statistical estimation based on advanced time series techniques. However, this process requires much more time and data, and we hope to continue it in our future work.
In parallel, we have analyzed dynamics of our major climate change variables -regional temperature, precipitation and sea level. This analysis was also based on advanced time series techniques, and we obtained some interesting results associated with five regional transportation hubs in our RRTN. As a matter of fact, we were able to capture long-run trends in these climate variables at each hub, and we plan to use these results further in our CGEM during simulation phase of our study.
So, in general at this point in time we have our model's architecture with some pieces already in place and with some others to be developed and estimated. We have climate change dynamics in terms of temperature, precipitation and sea level. Next step is to combine the two, expand our database and add some other climate change impacts and first of all frequency and magnitude of large weather events. In parallel, we are going to design user-friendly interface which will allow policy makers to easily use our CGEM without knowing all the model's details.