Dutch disease eh? : the oil boom and Canadian manufacturing

(1)

Dutch Disease Eh?

The Oil Boom and Canadian Manufacturing

Travis Westover

11799986

10 August, 2018

University of Amsterdam – Amsterdam School of Economics

MSc Economics – International Economics and Globalisation

Email: travis.westover@student.uva.nl

Supervisor: Cenkhan Sahin

Second Reader: Dirk Veestraeten

Word Count: 10,726

(2)

Statement of Originality

This document is written by Student Travis Westover who declares to take full responsibility for the contents of this document.

I declare that the text and the work presented in this document are original and that no sources other than those mentioned in the text and its references have been used in creating it.

The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.

(3)

1. Introduction

Canada is a very large and economically diverse country, with various industries

concentrated in different regions. The federal constitutional makeup of Canada creates a political backdrop that is relatively unique. Provinces have significant powers under the constitution, including complete control over the extractive resource industry, including the collection of royalties. The oil industry in Canada is concentrated primarily in the western provinces Alberta, Saskatchewan and British Columbia. Some additional oil production occurs on the east coast in Newfoundland and Labrador. This contrasts with the regional concentration of manufacturing located in the Windsor (Ontario) to Quebec City (Quebec) corridor in central Canada, in which the majority of the Canadian population resides. This regional variation creates a unique political situation and has resulted in political tension between regions in the past.1 The regional concentration of the oil industry in the west in contrast to the concentration of the manufacturing industry in central and eastern Canada magnifies this issue further. Many of the benefits of developing oil deposits in Alberta accrue locally, such as direct employment and royalty revenues, while some of the perceived drawbacks such as adverse manufacturing impacts accrue nationally, or in other regions. This fact combined with the fact that resource revenue falls under provincial government

jurisdiction rather than federally has made the impact of resource development a politically volatile issue. Frequently, one of the most discussed drawbacks is the notion that Canada suffers from Dutch disease.

Dutch disease, the term first coined by the Economist magazine in 1977 (2010) refers to the loss of competitiveness of a nation’s export manufacturing sector which is precipitated by an appreciation in the real exchange rate.2 This can occur due to large inflows of foreign aid in developing countries or, the focus of this thesis, due to resource booms. Large inflows of capital to develop the oil assets or the spike in the resulting income generated from the oil boom can both drive real exchange rate appreciation. Although the oil boom of 2002 – 2014 has now come to an end, with prices declining and exceptional returns are no longer being generated by the Canadian oil industry, the question of if, and to what extent Canada suffered from Dutch disease during the recent boom remains a salient one.

1_{This is further discussed in Section 3}

2_{While more focused on impacts of the supposed resource curse in developing nations, an interesting overview}

(5)

My research question evaluates to what extent Canada has suffered from Dutch disease effects by examining if the real exchange rate appreciation has been driven by increases in energy prices, and if so, whether this appreciation had a negative impact on manufacturing activity. I will also assess if these effects have been constant over time, or if they have varied as the Canadian economy has developed. As there was a significant commodity boom – particularly in energy prices – beginning in the early 2000s, this thesis will also seek to examine if there has been a structural shift in the drivers of the real exchange rate, and to attempt to quantify this impact if it is observed.

There is ample literature describing the theoretical framework around Dutch disease effects, specifically those caused by a resource boom. More recent literature has taken an empirical approach to analyzing these effects in the Canadian context, with mixed

conclusions. Studies by Beine et. al. (2012) and Shakeri et. al. (2012) conclude that Canada has suffered some negative effects of Dutch disease, while other papers by Naim & Tombe (2013) and Cross (2013) conclude Canadian manufacturers may have in fact enjoyed some

benefits from real exchange rate appreciation. It is clear there is no consensus on either the

directionality nor magnitude of Dutch disease effects on Canada. Consequently, there is room for further research. In particular, looking at manufacturing shipment data instead of

employment impacts in the manufacturing industry could help to separate Dutch disease effects from the general trend towards automation in the manufacturing industry. Using manufacturing shipments as a proxy for output may better capture the value added in

manufacturing, and this data has the benefit of being available on a higher frequency monthly basis, compared to only quarterly GDP data.

The empirical methodology in this paper will build on the work of Beine et. al (2012); however, with one significant change in explanatory variables: this paper will use data with a higher frequency (monthly in addition to quarterly) and use manufacturing shipments as a proxy for manufacturing activity. Employment, while easier to measure, is likely exposed to a secular decline over the period analyzed by Beine et. al., as the shift away from labour towards capital is particularly notable in the manufacturing sector. For robustness, quarterly data using manufacturing GDP is also used. The model is then extended to include a

difference-in-difference analysis, (as applied to Dutch disease and commodity boom

empirical analysis in papers by Vicente (2010) and Weber (2012)) to evaluate the impact of the commodity boom on both the real exchange rate and manufacturing activity. The difference-in-difference analysis will allow us to examine the impact of the commodity boom, and to address the second part of the research question i.e. if the data show a structural

(6)

change in the drivers of the real exchange rate and manufacturing activity. The most closely related work to this study is the 2012 paper of Beine et al. who examine the empirical relationship between real exchange rate appreciation and manufacturing employment, and while this paper builds on their work, it is extended to include different measures of

manufacturing activity and an additional difference-in-difference econometric specification to examine the impacts of the commodity boom.

This paper’s findings mirror that of the body of previous literature in that we do not reach a definitive conclusion regarding Dutch disease effects on Canada. While it is clear that the real exchange rate appreciated significantly, and we empirically substantiate that this appreciation was driven by increases in energy prices, this only confirms the first piece of the Dutch disease puzzle. The second piece – that this real exchange rate appreciation has had an adverse impact on manufacturing activity – is not conclusive. While the causal relationship between real exchange rate appreciation and manufacturing activity has been shown to be weakly negative in some model specifications, we cannot rule out that it is zero. These mixed findings appear to be consistent with some recent literature, suggesting that although there is evidence of commodity boom driven real exchange rate appreciation, Canadian

manufacturing may not have been adversely impacted, and in fact may have enjoyed potential benefits of an appreciated currency.

The remainder of the thesis is structured as follows – Section 2 provides an overview of the state of current literature and its relation to the topic in addition to presenting a critical discussion of previous research in this area. Section 3 discusses the background of the Canadian economy and energy industry. Section 4 discusses the data selection and

methodology for the empirical analysis, while Section 5 presents the results. Finally, Section 6 concludes the paper.

(7)

2. Literature Review

There is quite a large body of literature on the subject of Dutch disease, with several seminal papers on the subject being published in the early 1980s, when this phenomenon was first being studied. More recent research builds on these early works, with a variety of theoretical and empirical extensions. This section will discuss some noteworthy papers, including recent empirical studies of the situation as it pertains to Canada.

Papers by Corden & Neary (1982) and Corden (1984) first established a framework model for analyzing Dutch disease effects in the context of a booming resource sector. These papers developed a model with three sectors, a booming (resource) sector B, lagging

(manufacturing) sector L, and a non-tradeable (services) sector N, and assumed the country being analyzed was a small, open economy. Accordingly, the boom in the resource sector drives increases in real incomes in that sector, which Cordon and Neary (1982) break down into two effects, namely the resource movement effect and the spending effect. The resource movement effect describes the impact of the increasing marginal product of mobile

production factors employed in the resource sector driven by the resource boom, drawing these out of other sectors (both N and L) and necessitating adjustments across the economy, including appreciation in the real exchange rate. The spending effect describes the impact of increased real income due to the resource boom on the demand for goods in other sectors, increasing their prices, which can further drive appreciation of the real exchange rate.

Corden (1984) expands on this core model by introducing rigidity in wages, as

workers in the manufacturing sector will seek employment in the booming resource sector, or if they remain in the manufacturing sector they will demand higher wages. This mechanism results in unemployment in the manufacturing sector. Corden also expands on previous work by looking at dynamics of current account balances, with current account deficits being generated in anticipation of booms and the period following them in contrast to a surplus being run during the boom itself. The findings of Corden and Neary (1982) and Corden (1984) were further confirmed by Van Wijnbergen (1984) who, using a Salter-Swan model, also found higher transfers led to a shift out of the (manufacturing) tradeables sector, and real exchange rate appreciation.3

3_{The Salter-Swan model of a small, open economy which produces both tradeable and non-tradeable goods}

was originally developed in the 1950s, but a more recent evaluation and extension of this model is done by Devarajan, Lewis, & Robinson (1993)

(8)

While the general findings in the literature were the same, i.e. that resource booms would lead to real exchange rate appreciation and thus impact the competitiveness of the manufacturing tradeables sector, Fardmanesh (1991) argues differently. His research suggests that despite rising oil prices throughout the 1970s, oil exporters were still able to grow their manufacturing output. In his model of three non-oil sectors – agriculture and manufacturing (products of which are exported) and non-tradeables – Fardmanesh claims that the

agricultural sector, and not manufacturing sector, was the one that bore the brunt of Dutch disease effects in this time period. He argues this is a result of the adverse spending effects of an oil boom on the manufacturing sector being offset by what he calls the “world price effect” whereby manufactured goods increase relative to agricultural goods. Thus in

Fardmanesh’s model, manufacturing may not be impacted by an oil boom, or may even grow during the boom. It is the agricultural sector that suffers the most in this model.

While the theoretical framework has been in place for some years now, only more recently have events prompted researchers to take a more in depth empirical look at the situation in Canada. The Canadian oil industry underwent tremendous growth in the first decade of the 21st_{century, and this rise in the economic importance of oil exports prompted} an increase in research in this field. A critical contribution to the literature in this area is by Beine, Bos and Coulombe (2012) who conducted an empirical study to ascertain if Canada was indeed suffering from Dutch disease. Their approach sought to break down fluctuations in the real exchange rate into component parts, differentiating between real US dollar (USD) depreciation and real Canadian dollar (CAD) appreciation. This approach enabled them to analyze impacts of real exchange rate changes separately, as market structure and

international competition can vary across sectors. Beine et. al. find that approximately 35% of the decline in Canadian manufacturing employment was driven by real exchange rate appreciation, and conclude that Canada has suffered from Dutch disease effects.

A potential significant shortcoming in the methodology of Beine, Bos and Coulombe however is the fact they have used employment to measure economic activity in the

manufacturing sector instead of output. While employment is much easier to measure than output, this approach could be significantly flawed as the early years of this century have been marked by massive increases in industrial automation. It is possible that these declines in manufacturing employment, while notable and occurring simultaneously with real

exchange rate appreciation, could be driven by technological factors. My paper will build on the work of Beine et al. while attempting to address these concerns.

(9)

Further analysis on the issue of Dutch disease in Canada was conducted in 2012 by Shakeri and his allies, whereby they evaluated the impact on a wide range of industries and sub industries of the appreciation of the Canadian dollar. They found that out of the eighty sub-industries in their analysis, twenty-five may have been negatively impacted (representing approximately one quarter of total manufacturing production) with the most impacted

generally being smaller industries that are more labour intensive, such as apparel

manufacturers. While they diagnose a “mild” form of Dutch disease for Canada they also note that similar declines in manufacturing have been observed in the United States – a significant importer of oil –which significantly undercuts their argument (Shakeri, Gray, & Leonard, 2012).

A unique approach taken in 2013 in separate papers by both Philip Cross and Naim & Tombe looks at the advantages to Canadian manufacturers of an appreciated currency. In their interesting analysis of the exposure of Canadian manufacturers to imported intermediate inputs, Naim & Tombe note that compared to other countries, Canadian manufacturers import a much larger proportion of their intermediate inputs (Naim & Tombe, 2013). They further discuss that, dependent on the elasticity of export demand, a real appreciation may in fact be a net benefit to Canadian manufacturers, potentially dispelling the notion that Canada may suffer from Dutch disease. Cross takes a similar viewpoint, arguing that Canadian manufacturers have made use of a “natural hedge” by switching to newly cheaper imported inputs in response to an appreciating currency. Cross further argues that it may be high energy prices, and not an appreciating currency, that may be responsible for the headwinds faced by Canadian manufacturing in the post crisis recovery (Cross, 2013). These papers’ arguments suggest that though real exchange rate appreciation may be present, it may not necessarily have a negative impact on manufacturing activity.

An interesting extension of previous literature was the paper of Papyrakis & Raveh (2014) who sought to extend the econometric framework to analyze regional Dutch disease in Canada. They found that resource rich provinces experienced higher inflation, which was their proxy for the spending effects of Dutch disease. Interestingly, they also found that resource rich provinces have international manufacturing exports negatively impacted by mineral production (again consistent with Dutch disease effects) while domestic exports to the rest of Canada were not impacted.

Using a difference-in-difference model to evaluate the impacts of commodity booms, a 2012 journal article by Weber examined the impacts of the natural gas boom on the county level in three US states. He used “boom counties”, those that experienced a significant spike

(10)

in gas production, as his treatment group, and “non-boom counties” as his control group. He found that boom counties experienced a statistically significant increase in income, but also experienced a decline in manufacturing share of earnings, indicating a possible county level Dutch disease diagnosis (Weber, 2012). Using a similar empirical approach, Vicente (2010) examines impacts of an oil boom in Sao Tome and Principe.4

In summary, there has been extensive research on the Dutch disease phenomenon since the term was first described by the Economist in 1977. Many papers in the 1980s established the theoretical framework of Dutch disease, and more recent empirical analyses have built on this framework. As the literature pertains to Canada, there have been but a few papers that seek to answer a similar research question to the one posed in this thesis. While some empirical studies such as Beine et. al. and Shakeri et. al. have led to the conclusion that Canada has experienced some Dutch disease effects, the opposing viewpoints espoused by both Naim & Tombe as well as Cross show that perhaps an energy boom driven increase in the real exchange rate may not solely negative for all Canadian manufacturers.

This paper’s contribution to the literature will be to further explore the impact of Dutch disease on the Canadian economy, by building on Beine et. al.’s methodology. An extension to the model is then developed, using a difference-in-difference methodology to evaluate the impact of the recent commodity boom, and quantify any structural shifts to the drivers of the Canadian real exchange rate and manufacturing activity.

4_{While Vicente examines corruption and not specific Dutch disease effects, his use of difference-in-difference}

(11)

3. Background on Canada’s Economy and Energy Industry

This section provides an overview of the Canadian economy, and will focus on the

manufacturing and energy industries in particular. Canada is the seventh largest oil producer in the world and the second largest in the Americas behind the United States (OECD, 2018). Oil and gas extraction directly makes up between 5-6% of GDP of Canada over the period 2007-2014 (Statistics Canada, 2018) although this measure only includes upstream activities directly related to extraction of oil and gas, and not any downstream activity such as oil transport and refining, nor does this include any administrative impacts and trickle down effects to the finance industry or broader economy. The boom in development of the oil industry in Canada has been a relatively recent phenomenon, as oil has increased as a share of total Canadian commodity production to 46% in 2012 from only 18% in the fifteen years previously (Carney, Dutch Disease, 2012)

Figure 1: Source: Canadian Association of Petroleum Producers

While Canada has been an oil exporter for the latter half of the 20th century, in recent years the rise in oil prices has driven a boom in production, and thus exports, from 1,221 thousand barrels/day in 1997 to 3,367 thousand barrels/day in 2017 (Canadian Association of Petroleum Producers, 2018). This represents an increase of 175% in two decades. This massive increase in production is clearly illustrated in Figure 1 above, in which natural gas production is also included for illustrative purposes.5 The United States is nearly the

exclusive destination for Canadian crude, as in the period from 2008-2015 99% of these oil

5_{The boom related spike in oil exports, and the declining significance of gas relative to oil are both visible in}

(12)

and gas exports were sent to the United States (Statistics Canada, 2018). The origin of these exports is also geographically concentrated. From 2002-2015, over 75% of Canadian energy exports originated in the western province of Alberta, with this number rising to over 80% in 2016-2017 (Statistics Canada, 2018). As is illustrated in Figures 2 and 3, there is a clear and large spike in oil prices from 2002 onwards which fuelled this boom in Canadian oil

production and exports.

Figure 2 Source: DataStream

Figure 3 Source: DataStream

Responses to this boom varied across Canada. In the oil producing west, the boom was welcomed and the “Great Recession” that enveloped so much of the developing world was of little impact. However, in late 2012, the leader of the left leaning Official Opposition

6_{An additional and timely topic related to the oil industry in Canada is the recent re-emergence of the United}

States as a major producing power. The shift of the United States towards energy independence due to increases in US production has been the major catalyst for the decline of global oil prices in recent years. As the US is essentially the only export market for Canadian crude at this time, this development has significant long-term implications for the Canadian energy industry and the Canadian economy as a whole. However, such a discussion is beyond the scope of this thesis.

(13)

New Democratic Party, Thomas Mulclair, proclaimed in an interview that Canada was suffering from Dutch disease (Gollum, 2012). This enhanced the political divide between the oil producing West, and the manufacturing centred Eastern part of Canada.7_{As political} rhetoric around Dutch disease escalated, then Bank of Canada governor Mark Carney directly addressed the issue with a speech in late 2012, and touched on the issue again in 2013

(Carney, Canada Works, 2013). Carney provided arguments against Dutch disease, pointing out the quick recovery of Canada relative to other nations from the “Great Recession” as well as the strong fundamentals for continued elevated oil prices.8 He further argued that the decline of manufacturing was both global and secular, as similar trends were observed in the United States. (Carney, Dutch Disease, 2012). Carney also roundly dismissed the notion that the Bank of Canada should lean against increases in the exchange rate (using active measures to intervene in the foreign exchange market) when considering monetary policy decisions, and mentioned that interprovincial trade barriers were creating friction in adjusting to the commodity boom in Canada.9

As is clear from this brief overview, the oil industry and its impact on the broader Canadian economy, as well as the significant increases in oil prices and Canadian oil

production and exports from 2002 onwards clearly made Dutch disease a hot political issue in the 2012-2014 period. While the political rhetoric has subsided greatly with the decline in oil prices from their highs in recent years, evaluating the facts and impacts of the oil boom on the real exchange rate and the manufacturing industry in Canada remains a salient issue.

7_{It should be noted that Alberta has generally tended to be more conservative than much of the rest of Canada,}

while urban Eastern Canada tends to lean more liberal politically. Thus political griping between the West and East has a long history in Canada, and is certainly not a new development.

8_{Strong fundamentals at the time (which have since significantly deteriorated) as the rapid increase in crude oil}

production in the United States had yet to materialize

9_{Canadian provinces enjoy a great deal of economic sovereignty, particularly over development of natural}

(14)

4. Data, Sources and Empirical Methodology

This section will discuss the data, sources and reasoning behind these choices, as well as discuss the empirical methodology of this paper.

4.1 Data and Sources

This paper uses various data to evaluate the causal relationship between fluctuations in the real exchange rate and price levels, as well as the relationship between these

fluctuations in the real exchange rate and manufacturing activity.

Data used that did not require any transformation, calculation or re-indexing include the Real Effective Exchange rate, Energy CPI, CPI Excluding Energy, Crude Production and Manufacturing Shipments. The Real Effective Exchange Rate is defined as a price level based measure of the real effective exchange rate (indexed to 2010) and is used as a proxy for the real exchange rate. Energy CPI, defined as the energy portion of the CPI basket, is used to reflect energy price levels in Canada. CPI excluding energy is defined as the price level of all goods in the CPI basket excluding energy and is used to reflect price levels of non-energy items in Canada. Daily average crude production is simply the monthly average volume of crude oil produced per day in Canada. Manufacturing Shipments is defined as the dollar value of new manufacturing shipments, and is used as a proxy for manufacturing activity. Data that required some calculations and transformations before use are the interest rate differential and real adjusted manufacturing GDP. The interest rate differential is defined as the difference between the 3-month interest rate in Canada and the 3-month interest rate in the United States and is included in the models to reflect potential market expectations regarding exchange fluctuations.

The data used to evaluate manufacturing activity was the most complex to find. On a monthly basis, a number of indicators were available, including new shipments, which is used as a proxy for manufacturing activity. Inventories and employment were also available, but have not been used, as inventories are volatile and may not reflect current economic activity as well as new shipments, while employment was used in previous empirical studies.10_{Manufacturing gross domestic product data was only available on a nominal,} quarterly basis, and was discontinuous, requiring the use of two separate series. Both of these were adjusted to a real number, using the GDP deflator, which was re-indexed to match the base year of each of the nominal series. There is also a gap in the availability of this data of

10_{See (Beine, Bos, & Coulombe, 2012)}

(15)

one quarter. Due to these adjustments, this variable is referred to as real adjusted manufacturing GDP.

Due to the fact that manufacturing GDP data was only available on a quarterly basis, the analysis are presented with a dual approach to frequency. Initially, monthly data are used with manufacturing new orders being the variable of interest to measure manufacturing activity. Next, quarterly data were used, with the real adjusted manufacturing GDP data being used to measure manufacturing activity in the quarterly analysis.

Both monthly and quarterly data used cover the time period of 1981-2017.This time period was the maximum available for many data points, and covers many interesting developments in the Canadian and global economies including the rise of Canada as an energy superpower, the commodity boom beginning in the early 2000s, as well as the Great Recession. Summary statistics for all variables used are available in the Data Appendix.

Monthly data was available for nearly all variables analysed, with the exception of manufacturing gross domestic product. Data aggregator DataStream was used to source all data. Energy CPI and CPI for all items excluding energy were used to evaluate changes in energy and non-energy price levels respectively in Canada during this time period, with the original data from Statistics Canada. Daily average crude oil production in Canada was available on a monthly basis, as was the interest rate differential between Canada and the United States.

4.2 Empirical Methodology

The empirical analysis of whether Canada is suffering from Dutch disease effects will be conducted first using OLS, with two parts to this analysis and then extended by the use of the difference-in-difference specification. First, using ordinary least squares (OLS) regression analysis, the paper evaluates if increases in energy price levels drive appreciation in the real exchange rate, and second, analyses if real exchange rate appreciation drives decreases in manufacturing output. If there is indeed a causal effect between increases in energy prices and reduced output in manufacturing, we would expect the coefficient on energy prices in the first regression to be positive, and the coefficient on the real exchange rate in the second regression to be negative. The model is then extended using a difference-in-difference model to evaluate the impact of the commodity boom on price levels and manufacturing activity.11

11_{These are performed on both monthly and quarterly data to enhance robustness, as well as due to the fact that}

(16)

For the OLS analysis, we are primarily interested in the changes in these variables and not their nominal levels and consequently use a log difference to evaluate percentage

changes. All variables were loaded into Stata in nominal form, and the transformations to log differences were done using Stata. We have the following specification for our first

regression, of the real effective exchange rate against non-energy CPI, energy CPI, oil production and the interest rate differential:

∆"#$_% = '₍+ '₊∆,-. /$ /01234_%+ '₅∆/01234 ,-._%+ '₆∆,2781 -2987:;<90_% + '₌∆. ><??_%+ @

In this regression, we would expect to have a positive coefficient β2 on energy CPI, as this is the first piece of the Dutch disease puzzle. Next, we then evaluate the relationship between changes in manufacturing shipments and changes in the real effective exchange rate, non-energy CPI, and non-energy CPI using the specification:

∆AB0Cℎ<E_% = '₍+ '₊∆"#$_%+ '₅∆,-. /$ /01234_%+ '₆∆/01234 ,-._%+ @

We would expect a negative coefficient β1 on the real effective exchange rate in this regression to confirm the second piece of the Dutch disease puzzle – that manufacturing activity is adversely effected by real exchange rate appreciation.

In both of the above specifications, the sample was analysed in its entirety (across the whole 1981-2017 period) as well as split into two sub-periods, from 1981-2001 and 2002-2017. As the commodity boom began around 2002, and we are interested in examining if there is a change in the drivers of the real exchange rate and manufacturing activity, splitting the sample allows us to see if the relevant coefficients maintain the same sign, magnitude and significance between the sub-periods, or if we can observe a change in these parameters.

Given we are interested in not only evaluating the relationships between the real exchange rate and its drivers but also the changes in these relationships over time, the model was then extended to include the use of a difference-in-difference specification. To evaluate the impact of the commodity boom, we introduced a dummy variable for the boom period, Time. The boom was assumed to begin in 2002, with the Time dummy variable being a 0 before the boom (1981-2001) and a 1 during the boom (2002-2017). The control group in this model is non-energy CPI. A difference-in-difference variable was constructed, as the product

(17)

of the energy CPI (the treatment group) and the boom dummy variable, Time. The following specifications are used for the difference-in-difference model, with additional control

variables being added for the sake of robustness:

"#$_% = '₍+ '₊F<G1 + '₅/01234 ,-._%+ '₆(/01234 ,-. ∗ F<G1) + @ "#$% = '(+ '+F<G1 + '5/01234 ,-.%+ '6 /01234 ,-. ∗ F<G1 + 'K"#$%L6+ @ "#$_% = '₍+ '₊F<G1 + '₅/01234 ,-._%+ '₆ /01234 ,-. ∗ F<G1 + '₌,-. /$ /01234_%+ '_M,2781 -2987:;<90_%+ '_N. ><??_%+ @ "#$% = '(+ '+F<G1 + '5/01234 ,-.%+ '6 /01234 ,-. ∗ F<G1 + '₌,-. /$ /01234_%+ '_M,2781 -2987:;<90_%+ '_N. ><??_%+ '_K"#$_%L6 + @

In the regression results in Table 3 in the next section, the coefficient on this difference-in-difference variable β3 can be interpreted as the causal effect of the treatment, the boom, on the real exchange rate. We would expect β3 to be positive and significant if the boom had an impact on the appreciation of the real exchange rate.

As there was no adequate control for the real exchange rate, the same treatment and control groups were used for the evaluation of the impacts on both the real exchange rate, the first piece of the Dutch disease puzzle, and the second piece of the Dutch disease puzzle. Thus similar specifications were used to those above, with manufacturing shipments replacing the real exchange rate:

AB0Cℎ<E_%= '₍+ '₊F<G1 + '₅/01234 ,-._%+ '₆(/01234 ,-. ∗ F<G1) + @

AB0Cℎ<E_%= '₍+ '₊F<G1 + '₅/01234 ,-._%+ '₆ /01234 ,-. ∗ F<G1 + 'KAB0Cℎ<E%L6+ @

AB0Cℎ<E_% = '₍+ '₊F<G1 + '₅/01234 ,-._%+ '₆ /01234 ,-. ∗ F<G1 + '₌,-. /$ /01234_%+ '_M,2781 -2987:;<90_%+ '_N. ><??_%+ @

(18)

AB0Cℎ<E_%= '₍+ '₊F<G1 + '₅/01234 ,-._%+ '₆ /01234 ,-. ∗ F<G1 + '₌,-. /$ /01234_%+ '_M,2781 -2987:;<90_%+ '_N. ><??_% + '_KAB0Cℎ<E_%L6+ @

Much as with the first difference-in-difference specifications, several models were used, with additional control variables being added. In the regression results in Table 4 in the next section, the coefficient on the difference-in-difference variable β3 can be interpreted as the causal effect of the boom on the manufacturing shipments. We would expect β3 to be negative if the boom had an adverse impact on manufacturing activity.

Following the monthly analysis, quarterly data was then analysed. This is done for the sake of robustness, as well as to provide the opportunity to evaluate a different measure of manufacturing activity, real adjusted manufacturing GDP, as this was only available on a quarterly basis. All specifications are as in the monthly section, although with manufacturing shipments being replaced by real adjusted manufacturing GDP.

(19)

5. Empirical Results

This section will discuss the empirical results and their implications. As GDP data was only available on a quarterly basis, regression results have been divided into two sections. First, we analyse the monthly data using manufacturing shipments as a proxy for manufacturing activity. This is then expanded to look at the difference-in-difference model, also on a

monthly basis. Next, we will look at the results using quarterly data, using real manufacturing GDP as the measure of manufacturing activity, as well as for robustness, to see if the

relationships examined at a monthly level hold at the quarterly time horizon as well.

5.1 Monthly Regression Results

This section will explain the monthly regression results, first using basic OLS with both the log-difference of the real exchange rate followed by the log-difference of

manufacturing shipments being used as dependent variables. The model is then extended, using difference-in-difference to examine the impacts of the structural shift in energy price trends from 2002 onwards.12

5.1.1 Real Exchange Rate as Dependent Variable

The first regression analysed monthly data from 1981-2017, with Equation 1 examining the whole period, and Equation 2 and 3 breaking the longer period down into 1981-2001 and 2002-2017. The results are detailed in Table 1 below. The choice of this break coincides with the beginning of the commodity super cycle in 2002, with tremendous increases in the price of oil and other energy commodities, as well as a significant increase in the production and export of these commodities in Canada during this timeframe, as

explained in Section 3 above.

12_{Initially, Augmented Dickey-Fuller tests, including no trend and 12 lags were run on the log-differenced}

variables to ensure that all were stationary and did not have a unit root; the MacKinnon approximate p-value for all variables was close to zero indicating that we can reject H0 at the 99% significance level and that the

(20)

Table 1

(1) (2) (3)

1981-2017 1981-2001 2002-2017

VARIABLES Δ Real Effective

Exchange Rate Δ Real Effective Exchange Rate Δ Real Effective Exchange Rate

Δ CPI Excluding Energy 0.749*** 0.964*** 0.620

(0.226) (0.218) (0.532)

Δ Energy CPI 0.109*** 0.0641* 0.119**

(0.0303) (0.0380) (0.0484)

Δ Crude Production -0.0239 -0.00485 -0.0479

(0.0165) (0.0172) (0.0312)

Δ Interest Rate Diff -0.00244 -0.00500*** 0.00310

(0.00158) (0.00150) (0.00347)

Constant -0.00254*** -0.00407*** -0.000988

(0.000932) (0.00104) (0.00159)

Observations 343 197 146

R-squared 0.089 0.163 0.089

Standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1

Some interesting results can be noted in Equation 1, where the coefficients are positive on both CPI excluding energy and energy CPI, and strongly significant. It is

interesting to note that the coefficient of non-energy CPI is much larger, with a coefficient of 0.749, suggesting non-energy CPI was the main driver of the changes in the real exchange rate across the whole 1981-2017 period as opposed to changes in energy prices.

When we divide the period into two sub periods, we get somewhat different results. Equation 2 looks at the 1981-2001 period, and ex-energy CPI is clearly the main driver of real exchange rate fluctuations in this time period, with a strongly significant and large coefficient of 0.964, with energy CPI only being weakly significant with a coefficient of 0.064. However, the results change drastically when we look at the 2002-2017 period in Equation 3. Here, the coefficient on non-energy CPI, while remaining positive as we would expect, is not statistically significant. However, the energy CPI coefficient is 0.119 and statistically significant at the 95% level. This increase in the coefficient and its significance from Equation 2 suggests that there was a structural shift in the data around 2002, which coincides with a period of significantly rising energy prices and growth in oil production in Canada.

We can conclude from Equations 1-3 that there is a link between the fluctuations of price levels and the real exchange rate. However, crude oil production, with a counter

(21)

intuitive negative coefficient across the entire range as well as the two sub-periods, is not statistically significant. It is clear that the real exchange rate responds to price level changes more than oil production changes. Additionally, the change in interest differential between Canada and the US was not significant over the entire 1981-2001 period (and very close to zero), although the coefficient for the earlier period was -0.005, and significant.

Overall, from Equations 1-3, we can conclude that changes in energy CPI have a positive coefficient and that the magnitude and significance of this relationship increased in the second period. The first condition of Dutch disease, that changes in the real interest rate are driven by increases in energy prices, appears to be confirmed by the data.

5.1.2 Manufacturing Shipments as Dependent Variable

Next, we now need to look at the empirical relationship between manufacturing “output” and the real exchange rate. This regression uses manufacturing shipments as a proxy for manufacturing output, and the results are visible in Table 2. Equation 4 looks at the entire 1981-2017 period, while as above, results were broken down into two sub-periods, with Equation 5 examining the 1981-2001 period, and Equation 6 looking at the 2002-2017 period. In this regression, we find some intriguing results that point to a structural shift in drivers of manufacturing shipments between the two sub periods.

As can be seen in Table 2 below, the real effective exchange rate has a negative coefficient that is weakly significant across the whole 1981-2017 period, in Equation 4. This coefficient remains negative in the sub periods (Equations 5 and 6), however it is not

statistically significant. This would appear to support the claim that real exchange rate appreciation has a negative causal relationship with manufacturing.

(22)

Table 2

(4) (5) (6)

1981-2017 1981-2001 2002-2017

VARIABLES Δ Manufacturing

Shipments Δ Manufacturing Shipments Δ Manufacturing Shipments

Δ Real Eff. FX Rate -0.128* -0.145 -0.111

(0.0673) (0.112) (0.0840)

Δ CPI Excluding Energy -0.354 -0.608 0.0207

(0.295) (0.371) (0.543) Δ Energy CPI 0.166*** 0.0415 0.233*** (0.0368) (0.0609) (0.0459) Constant 0.00322*** 0.00570*** 0.000702 (0.00113) (0.00162) (0.00157) Observations 443 251 192 R-squared 0.050 0.024 0.121

The coefficients for both non-energy and energy CPI also further support the notion of a significant shift in drivers of Canadian manufacturing activity over the time period examined. The coefficient for non-energy CPI is not statistically significant across Equations 4-6, however the results are quite different for the energy CPI. It is positive, and highly significant for the entire period as can be seen in Equation 4. This does not tell the entire story however, as in the early period in Equation 5, it is small and not statistically significant. In Equation 6, it is highly significant and increases. This is a rather compelling result, and may suggest that increase in energy prices from 2002 onwards in fact increased

manufacturing activity.

Overall, results of this regression are somewhat unclear. The negative coefficient on real exchange rate changes during the entire period is only weakly significant (at the 90% level), but would appear to confirm that real exchange rate appreciation has negatively impacted manufacturing shipments during the whole 1981-2017 period. However, the

significance and sign on energy CPI changes is the most intriguing result and may suggest the opposite of a Dutch disease diagnosis.

(23)

5.1.3 Difference-In-Difference

Given the results in the above regressions, the divergence in energy prices and their impact on the real exchange rate merits further analysis. To further analyze the causal relationship between changes in energy prices and the real exchange rate, the model was extended to include a difference-in-difference analysis, with a dummy variable for the boom time (from 2002 onward being 1, otherwise 0), to look at the specific impact of the

divergence of energy prices from the CPI excluding energy, as is visible in Figure 4. The choice of 2002 for the beginning of the boom mirrors the division of the OLS results above into the two sub periods, pre and during the boom. As is clear from Figure 4, there is a common trend between energy CPI (CP_NRG) and non-energy CPI

(CPI_EX_NRG) up to approximately 2002, when the oil boom began and a clear divergence is visible of the energy CPI. Manufacturing shipments (scaled down by a factor of one million to be easier to compare to the levels of the other variables on the same graph) also exhibits a common trend up to 2002.13

Figure 4

13_{The common trend assumption is critical for the validity of the estimates in the difference-in-difference}

model, as it is assumed that the common trend would have continued if not for the treatment (in this case the boom). For a more detailed discussion of difference-in-difference methodology see (Angrist & Pischke, 2009)

0 50 1 00 1 50 2 00 1980m1 1990m1 2000m1 2010m1 2020m1 month CN_RFX CPI_EX_NRG CP_NRG MAN_SHP_scaled

(24)

This regression uses the CPI excluding energy as the control group, and the energy CPI as the treatment group, with the boom itself being the treatment. The results are visible in Table 3 below.

In Equation 1 in Table 3, we can see that the DiD estimator, the coefficient on the (boom*Energy CPI) variable in the model is positive, and highly significant. This is the causal effect of the boom on the real exchange rate, and implies that the boom led to an appreciation of the currency in real terms. In Equation 3, we can see that even with additional control variables added, the coefficient is still positive and significant. These results would appear to support the first piece of the Dutch disease puzzle, that the commodity boom from 2002 onward led to appreciation in the Canadian dollar.

In Equations 2 and 4, we add a lagged value of the dependent variable to the regression to further enhance the robustness of the analysis. As we can see, the DiD coefficient is much smaller, however, it is still positive and significant. Using a lagged

variable specification in a DiD model tends to bias the DiD estimate downwards, whereas the model specification in Equations 1 and 3 tends to be biased upwards (Angrist & Pischke, 2009). Using both specifications can provide an upper and lower bound of the likely range of

Table 3

(1) (2) (3) (4)

VARIABLES Real FX Rate Real FX Rate Real FX Rate Real FX Rate

Time -61.65*** -1.473 -61.58*** -7.803*** (4.418) (1.881) (5.750) (2.729) Energy CPI -0.427*** -0.0557*** -0.165*** -0.0662** (0.0332) (0.0137) (0.0617) (0.0263) DiD Estimator 0.670*** 0.0407** 0.663*** 0.108*** (0.0422) (0.0187) (0.0594) (0.0283)

CPI Excluding Energy 0.0666 0.0326

(0.0640) (0.0270)

Crude Production -0.0132*** -0.00268***

(0.00125) (0.000575)

Interest Rate Diff 1.588*** -0.00626

(0.211) (0.0957)

3rd Lag of Real FX Rate 0.937*** 0.890***

(0.0165) (0.0200)

Constant 119.1*** 9.464*** 113.7*** 16.26***

(2.488) (2.156) (2.616) (2.499)

Observations 444 441 444 441

(25)

the DiD estimator (Angrist & Pischke, 2009, p. 184). In this way, we can see from the above results that the causal impact of the boom on the appreciation of the real exchange rate is likely positive, and statistically significant. In Equation 4, we add additional variables, and the DiD estimator remains positive and highly significant.

While the results in Table 3 above support the earlier findings of a causal link between increases in energy prices and appreciation in the real exchange rate, the results in Table 3 are more robust than the findings using OLS in Table 1, as we have much higher R-squared values.14 Overall, the results in Table 3 are consistent with, and build upon the results in Table 1 and show a positive causal relationship between increases in energy prices and the real exchange rate. The first condition of Dutch disease appears to be empirically valid using both OLS and difference-in-difference methodology.

Next, we analyse the impact of the boom on manufacturing activity, again using manufacturing shipments as a proxy for manufacturing activity, as in Table 2 above. These results can be seen below in Table 4. As discussed in the methodology section, there was not an adequate control group available for the real exchange rate. As such, the coefficient on the DiD estimator is the impact of the energy boom, or increase in energy CPI on manufacturing shipment levels.

As is visible in Equation 5, there is a negative and significant DiD coefficient, which would suggest that the boom had a negative treatment effect on manufacturing activity. In Equation 7, the sign and significance of this relationship hold when we add additional independent variables. However, this relationship does not appear to withstand additional rigour of the specification in Equations 6 and 8, which include the third lagged value of the dependent variable. This lag was chosen as a one quarter lag seemed most appropriate, and is easily comparable to the quarterly results in the next section. The DiD estimator is, while negative, no longer statistically significant.

14_{While we are chiefly interested in the coefficient measuring the causal impact of the boom on the real}

exchange rate, other interesting results include a somewhat counterintuitive and very small, although significant, negative coefficient on crude production.

(26)

Table 4 (5) (6) (7) (8) VARIABLES Manufacturing Shipments Manufacturing Shipments Manufacturing Shipments Manufacturing Shipments

Time 5.728e+07*** 948,100 5.411e+07*** 784,399

(2.231e+06) (1.319e+06) (3.263e+06) (1.695e+06)

Energy CPI 622,842*** 25,894** 499,628*** 18,475

(16,779) (12,757) (34,998) (17,097)

DiD Estimator -553,335*** -14,043 -515,772*** -9,514

(21,329) (12,721) (33,731) (16,988)

CPI Excluding Energy 4,063*** 73.24

(707.9) (288.1)

Crude Production -3,267 6,455

(36,336) (14,199)

Interest Rate Diff -811,831*** -201,810***

(119,540) (48,400)

3rd Lag of Man-Ship 0.960*** 0.934***

(0.0175) (0.0192)

Constant -1.823e+07*** -482,571 -1.421e+07*** 491,453

(1.257e+06) (576,465) (1.484e+06) (679,598)

R-squared 0.923 0.990 0.939 0.991

The results in Equations 5 and 7 would seem to support the negative causal link between the boom and a decrease in the level of manufacturing activity. However, the lack of a statistically significant coefficient on the DiD estimator in Equations 6 and 8 suggests that no causal link may exist. With zero as an upper bound in Equations 6 and 8, and a negative significant coefficient in Equations 5 and 7, it is possible there may be a negative causal effect, but we cannot determine this with certainty.

Overall, the difference-in-difference extension to the model confirms and builds upon the conclusions from the OLS model with respect to the positive causal relationship between the energy boom and the real exchange rate. It is clear that there was a structural change in the drivers of the real exchange rate during the commodity boom, where energy prices became a more significant driver of these changes from 2002 onward, during the boom. The results on the second piece of the Dutch disease puzzle from the difference-in-difference are somewhat more opaque. While it appears there may be a negative relationship between the boom and manufacturing, it cannot be empirically proven with these data. As with OLS, the second condition for Dutch disease is unclear, and the most interesting result remains the

(27)

clear divergence between energy prices and price levels in the rest of the economy during the boom.

5.2 Quarterly Regression Results

In the interest of robustness, quarterly data was also analysed. Additionally, as manufacturing GDP data was only available quarterly, it is used instead of manufacturing shipments as a proxy for manufacturing activity in the regressions below. As with monthly data, first basic OLS was used, followed by an extension with the application of a difference-in-difference model.15

5.2.1 Real Exchange Rate as Dependent Variable

The first regression in this section analyzed the same data as Equations 1-3 in Table 1 above, but on a quarterly basis. As detailed in Table 5 below, we can see that the coefficient on non-energy CPI is positive, and weakly significant across the whole 1981-2017 period (Equation 1). Interestingly, we can see a structural break when we examine the two sub periods in Equations 2 and 3 – non-energy CPI is strongly significant in the earlier period, but not significant in the later period. This would appear to suggest that non-energy price levels were less important for driving changes in the real exchange rate from 2002 onwards.

15_{Initially, Augmented Dickey-Fuller tests, including no trend and 4 lags were run on the log-differenced}

variables to ensure that all were stationary and did not have a unit root; the MacKinnon approximate p-value for all variables was close to zero indicating that we can reject H0 at the 99% significance level and that the variables are stationary.

(28)

Table 5

(1) (2) (3)

1981-2017 1981-2001 2002-2017

VARIABLES Δ Real Effective

Exchange Rate Δ Real Effective Exchange Rate Δ Real Effective Exchange Rate

Δ CPI Excluding Energy 0.661* 1.372*** -1.436

(0.366) (0.302) (1.646)

Δ Energy CPI 0.164*** 0.00309 0.255**

(0.0585) (0.0703) (0.0949)

Δ Crude Production -0.00161 -0.0533 0.0774

(0.0535) (0.0587) (0.0926)

Δ Interest Rate Diff -0.00227 -0.00730* 0.00400

(0.00446) (0.00376) (0.00979)

Constant -0.00723** -0.0160*** 0.00495

(0.00358) (0.00361) (0.00823)

Observations 112 64 48

R-squared 0.117 0.307 0.173

This becomes more interesting, as we can see the coefficient on energy CPI is strongly significant, and positive across the entire period, and while not significant in Equation 2, it is significant in Equation 3. This would suggest a shift in drivers of real

exchange rate changes from non-energy prices pre-2002 to energy prices from 2002 onwards. As the coefficients on energy CPI are positive, the data appear to confirm the first phase of Dutch disease – a causal link between changes in energy prices and the real exchange rate.

5.2.2 Real Adjusted Manufacturing GDP as Dependent Variable

While the results above support the Dutch disease hypothesis, the results below do not provide much value in establishing any causal link between real exchange rate fluctuations and manufacturing output, in this case measured as real adjusted manufacturing GDP. The coefficient on the real effective exchange rate, energy CPI and crude production are all not statistically significant. As can be seen below in Table 6, the only significant coefficient is on non-energy CPI in Equation 4, for the entire 1981-2017 period and 5 for the pre-boom period of 1981-2001. Interestingly, these coefficients are negative. This may be due to the

exceptionally high inflation experienced in the 1980s, but is of little value for the purposes of this thesis.

(29)

Table 6

(4) (5) (6)

1981-2017 1981-2001 2002-2017

VARIABLES Δ Real ADJ

Manufacturing GDP

Δ Real ADJ Manufacturing GDP

Δ Real Eff. FX Rate -0.0285 0.0622 0.0225

(0.0680) (0.129) (0.0729)

Δ CPI Excluding Energy -1.277*** -1.947*** 0.764

(0.271) (0.342) (0.779) Δ Energy CPI -0.0279 -0.0432 -0.0479 (0.0423) (0.0714) (0.0474) Δ Crude Production 0.0430 0.0473 0.0334 (0.0356) (0.0622) (0.0385) Constant 0.00591** 0.0159*** -0.00795** (0.00242) (0.00371) (0.00383) Observations 145 83 62 R-squared 0.169 0.353 0.049

Overall, looking at quarterly OLS data, it is not possible to claim that any causal link exists, as the results of Equations 4-6 in Table 6 are generally not significant. It is interesting to note the shift in the main driver of the change in the real exchange rate however, as in Equation 2 in Table 5 it is clearly non-energy prices, while energy prices are highly

significant in Equation 3. These results are generally in line with the monthly data in showing a clear structural shift to the driver of the real exchange rate, but do not show evidence of a link between this appreciation and a decrease in manufacturing activity. Additionally, the constructed variable Real Adjusted Manufacturing GDP may be flawed, as construction of it required the merging of two separate, discontinuous series. As a result, unfortunately, we cannot draw any causal inferences from the regressions in Table 6.

(30)

5.2.3 Difference-In-Difference

As with the monthly data, a difference-in-difference extension to the analysis was performed for the quarterly data. In Table 7, we can see the results are broadly similar to the monthly data in Equations 1 and 3, with the difference-in-difference estimator being positive and highly significant; in fact, the values of the difference-in-difference coefficients in these Equations are very similar to those from the monthly data. This would suggest that the treatment effect of the boom is positive. However, in Equations 2 and 4, when the first lagged value of the real exchange rate is included in the regression, the coefficient for the difference-in-difference estimator is no longer significantly different from zero. Much like the monthly data, we have established a range of possible values for the difference-in-difference estimator, however, in contrast to the monthly results and given that the lower boundary is zero, we cannot conclusively state that the boom led to an appreciation of the real exchange rate.

Table 7

(1) (2) (3) (4)

VARIABLES Real FX Rate Real FX Rate Real FX Rate Real FX Rate

Time -62.47*** -0.138 -69.54*** -4.544 (7.714) (2.961) (10.05) (4.691) Energy CPI -0.431*** -0.0496** -0.207* -0.0533 (0.0577) (0.0214) (0.107) (0.0432) DiD Estimator 0.678*** 0.0279 0.743*** 0.0742 (0.0736) (0.0294) (0.104) (0.0486)

CPI Excluding Energy 0.137 0.0156

(0.112) (0.0448)

Crude Production -0.00527*** -0.000603*

(0.000754) (0.000344)

Interest Rate Diff 1.497*** -0.0125

(0.352) (0.151)

1st_{Lag of Real FX Rate} _0.952*** _0.916***

(0.0258) (0.0339)

Constant 119.4*** 7.697** 115.5*** 12.89***

(4.319) (3.383) (4.447) (4.264)

R-squared 0.396 0.944 0.652 0.946

(31)

Moving on to the next and final regression, we can analyse the impact of the commodity boom on the real adjusted manufacturing GDP. These results are rather compelling and are visible in Table 8 below. As in the monthly regressions, Equation 5 includes simply the difference-in-difference variable, Equation 6 the first lagged value of real adjusted manufacturing GDP, Equation 7 includes the difference-in-difference estimator and control variables, while Equation 8 includes both the lagged value and control variables.

Table 8

(5) (6) (7) (8)

VARIABLES Real ADJ

Man-GDP Real ADJ Man-GDP Real ADJ Man-GDP Real ADJ Man-GDP Time 10,147*** 1,315 29,254*** 622.2 (3,493) (889.6) (5,339) (1,499) Energy CPI 40.95 18.71*** 269.7*** 1.149 (26.02) (6.618) (56.87) (15.72) DiD Estimator -77.22** -18.08** -278.9*** -8.930 (33.24) (8.437) (55.26) (15.32) Crude Production 1.422*** 0.145 (0.401) (0.104)

CPI Excluding Energy -324.2*** -5.920

(59.70) (16.55)

Interest Rate Diff -357.4* -213.1***

(187.8) (47.08)

1st Lag of Real Man-GDP 0.922*** 0.913***

(0.0205) (0.0212)

Constant 35,739*** 1,586* 37,754*** 3,342***

(1,948) (875.3) (2,366) (967.5)

R-squared 0.130 0.945 0.306 0.954

The difference-in-difference estimator has a negative, and statistically significant coefficient in Equations 5-7. This implies that the commodity boom had a negative impact on manufacturing activity, which would be supportive of the Dutch disease hypothesis. This is rather intriguing as, compared to the OLS results, we have statistically significant outcomes. The most compelling result in this regression is in Equation 6, with a high R-squared value of 0.945, and a negative difference-in-difference estimator valid at the 95% significance level. Although the coefficient on the difference-in-difference estimator is not statistically

significant in Equation 8, these results seem to lend support to a negative relationship between the commodity boom and real manufacturing GDP.

(32)

6. Conclusions

The research question this paper sought to answer is whether Canada has suffered from Dutch disease, and if these impacts were constant or had changed over time. To

determine whether Canadian manufacturing has suffered from Dutch disease effects due to the oil boom, we have established two conditions which had to be met: appreciation in the real exchange rate must be driven by increases in energy prices, and the appreciation of the real exchange rate must have a negative impact on manufacturing activity. To determine whether these impacts were constant or varied over time, we broke down the period analyzed into sub periods, as well as applied difference-in-difference

methodology.

This paper was more successful in analyzing the first condition of Dutch disease than the second. From the empirical analysis, we can conclude that energy CPI was a

significant driver of real exchange rate appreciation. Using both monthly and quarterly data, there appears to be a positive causal relationship between energy prices and the real exchange rate. The energy price driven appreciation of the real exchange rate suggests that the first piece of the Dutch disease diagnosis appears to be valid in Canada across the period examined. However, the second piece – what impact this appreciation had on manufacturing activity – is at best inconclusive. It has been shown to be negative, although with generally weak or zero significance, particularly in the OLS specification of the model. Using the difference-in-difference specification, the causal impact of real exchange rate appreciation would appear to be negative, although again we cannot conclusively rule out that it is zero.

Delving further into these results, we can also conclude that the relationship between the real exchange rate and energy prices was not constant over time, but underwent a structural shift beginning around 2002. This shift coincides with the commodity boom, during which energy price levels became a more significant driver of real exchange rate appreciation than in the past. Using the difference-in-difference specification, we can also conclude that the boom period resulted in a higher appreciation of the real exchange rate than would otherwise have been the case. We can also see a potential weakly negative relationship between the energy boom and real manufacturing GDP, although it is

important to reiterate that in most model specifications, it was not statistically significant. Overall, the first condition of Dutch disease is supported by the empirical facts while the second condition is not conclusively supported. Perhaps Canadian manufacturers

(33)

benefitted from the real exchange rate appreciation, enjoying the benefits of the “natural hedge” (Cross, 2013) by switching to imported inputs to take advantage of the

strengthening currency. Given Canadian manufacturers tend to have a much higher proportion of imported inputs than in other countries (Naim & Tombe, 2013) it is perhaps overly simplistic to simply claim that real exchange rate appreciation would or should have a negative impact on Canadian manufacturers. Additionally, the period analyzed was not a favourable one for manufacturing elsewhere, as declines were observed in the United States (Shakeri, Gray, & Leonard, 2012) – Canada’s chief trading partner.

Potential for further research into the impacts of the commodity boom on Canada is vast. The model in this paper does not include agriculture, which may have been more adversely impacted than manufacturing during the oil boom. Future research could expand on that of Cross, with a much more complex multi-sector model broken down at the monthly level versus the typical quarterly or annual data used in previous research. Another compelling area for further research is the phenomenon of regional Dutch

disease, as pioneered by Papyrakis & Raveh in 2014. Given the heterogeneous makeup of the Canadian economy, and the country’s vast size and regional concentration of certain industries, further analaysis of regional versus international impacts of Dutch disease could be explored.

(34)

Bibliography

Angrist, J. D., & Pischke, J.-S. (2009). Mostly Harmless Econometrics: An Empiricists's

Companion. Princeton, NJ, United States: Princeton University Press.

Beine, M., Bos, C. S., & Coulombe, S. (2012). Does the Canadian Economy Suffer from Dutch disease? Resource and Energy Economics , 468-492.

Canadian Association of Petroleum Producers. (2018, May 22). CAPP Statistical Hanbook. Retrieved May 22, 2018, from CAPP Publications and Statistics:

https://www.capp.ca/publications-and-statistics/statistics/statistical-handbook

Carbone, J. C., & McKenzie, K. J. (2015, June). Going Dutch? The Impact of Oil Price Shocks on the Canadian Economy. University of Calgary Working Paper . Calgary, Alberta, Canada: University of Calgary.

Carney, M. (2013, May 21). Canada Works. Remarks by Mark Carney . Montreal, Quebec, Canada: Bank of Canada.

Carney, M. (2012, September 7). Dutch Disease. Remarks by Mark Carney . Calgary, Alberta, Canada: Bank of Canada.

Corden, W. (1984). Booming Sector and Dutch Disease Economics: Survey and Consolidation. Oxford Economic Papers , 36 (3), 359-380.

Corden, W. M., & Neary, J. P. (182). Booming Sector and De-Industrialisation in a Small Open Economy. The Economic Journal , 825-848.

Cross, P. (2013). Dutch Disease, Canadian Cure: How Manufacturers Adapted to the Higher

Dollar. Ottawa: Macdonald-Laurier Institute.

Devarajan, S., Lewis, J. D., & Robinson, S. (1993). External Shocks, Purchasing Power Parity, and the Equilibrium Real Exchange Rate. The World Bank Economic Review , 45-63. Fardmanesh, M. (1991). Terms of trade shocks and structural adjustment in a small open economy. Journal of Development Economics , 339-353.

Gollum, M. (2012, May 19). Is Canada Suffering From Dutch Disease? Retrieved June 24, 2018, from CBC News: https://www.cbc.ca/news/canada/is-canada-suffering-from-dutch-disease-1.1140014

Jbir, R. (2013). The LInk Between Oil Price and Dutch Disease. Energy Sources, Part B , 391-397.

Naim, W., & Tombe, T. (2013). Appreciate the Appreciation: Imported Inputs and the

Concern Over Dutch Disease. University of Calgary , School of Public Policy. Calgary:

(35)

OECD. (2018, June 10). Crude Oil Production. Retrieved June 10, 2018, from OECD Data: https://data.oecd.org/energy/crude-oil-production.htm

Papyrakis, E., & Raveh, O. (2014). An Empirical Analysis of a Regional Dutch Disease: The Case of Canada. Enviromental Resource Economics , 179-198.

Shakeri, M., Gray, R. S., & Leonard, J. (2012). Dutch Disease or Failure to Compete? A

Diagnosis of Canada's Maufacturing Woes. Montreal: Institute for Research on Public

Policy.

Statistics Canada. (2018, May 22). Table 379-0029 - Gross domestic product (GDP) at basic

prices, by North American Industry Classification System (NAICS), annual (dollars).

Retrieved May 22, 2018, from CANSIM:

https://www150.statcan.gc.ca/n1/en/type/data?MM=1

Statistics Canada. (2018, May 20). Trade Data Online. Retrieved May 20, 2018, from Trade Data Online: http://www.ic.gc.ca/eic/site/tdo-dcd.nsf/eng/Home

The Economist. (2010, September 9). It's only natural. The Economist , p. 6.

van Wijnbergen, S. (1984). Inflation, Employment and the Dutch Disease in Oil-Exporting Countries: A Short-Run Disequilibrium Analysis. The Quarterly Journal of Economics , 233-250.

Vincente, P. C. (2010). Does oil corrupt? Evidence from a natural experiement in West Africa. Journal of Development Economics (92), 28-38.

Weber, J. G. (2012). The effects of a natural gas boom on employment and income in Colorado, Texas and Wyoming. Energy Economics , 34, 1580-1588.

(36)

Data Appendix

Monthly Data Summary Statistics

Variable Obs Mean Std. Dev. Min Max

Canadian Real Effective Exchange Rate 444 89.16 8.94 71.56 106.23

CPI Excluding Energy 444 93.53 21.72 47.50 129.20

Energy CPI 444 101.79 36.08 41.90 173.00

Crude Production 444 2,188 751 955 4,230

Manufacturing Shipments 444 36,800,000 12,700,000 15,000,000 55,900,000

Canada 3m 444 5.37 4.39 0.16 20.90

US 3m 444 4.21 3.45 -0.01 15.92

Δ Canadian Real Effective Exchange Rate 443 -0.01% 1.40% -8.60% 5.36%

Δ CPI Excluding Energy 443 0.23% 0.31% -0.81% 2.59%

Δ Energy CPI 443 0.30% 2.51% -12.09% 9.59%

Δ Crude Production 443 0.26% 4.39% -24.00% 20.39%

Δ Manufacturing Shipments 443 0.29% 1.95% -9.65% 6.38%

Interest Rate Differential 444 1.16 1.61 -2.40 5.65

Δ Interest Differential 343 -0.91% 46.84% -188.27% 331.42%

Period 444 223 128 1 444

Time Dummy (2002 or later = 1) 444 0 0 0 1

(37)

Quarterly Data Summary Statistics

Variable Obs Mean Std. Dev. Min Max

Canadian Real Effective Exchange Rate 148 89.16 8.92 71.92 104.94

CPI Excluding Energy 148 93.54 21.76 48.00 129.00

Energy CPI 148 101.79 36.10 42.70 172.30

Crude Production 148 6,527 2,224 3,467 12,161

Canada 3m 148 5.35 4.37 0.19 19.35

US 3m 148 4.16 3.41 -0.01 14.12

Manufacturing Adjusted GDP 147 39,654 3,354 32,451 49,077

Δ Canadian Real Effective Exchange Rate 147 -0.02% 2.47% -10.22% 6.20%

Δ CPI Excluding Energy 147 0.67% 0.60% -0.46% 3.20%

Δ Energy CPI 147 0.88% 4.05% -20.97% 11.74%

Δ Crude Production 147 0.75% 4.48% -20.16% 19.46%

Interest Rate Differential 148 1.18 1.63 -2.31 5.55

Δ Interest Differential 112 -0.03 0.53 -1.66 1.26

Δ Manufacturing Adjusted GDP 145 -0.26% 2.05% -6.64% 3.67%

Period 148 75 43 1 148

Time Dummy (2002 or later = 1) 148 0.43 0.50 0 1.00

Dutch disease eh? : the oil boom and Canadian manufacturing

Dutch Disease Eh?

The Oil Boom and Canadian Manufacturing

Travis Westover

11799986

10 August, 2018

University of Amsterdam – Amsterdam School of Economics

MSc Economics – International Economics and Globalisation

Email: travis.westover@student.uva.nl

Supervisor: Cenkhan Sahin

Second Reader: Dirk Veestraeten

Word Count: 10,726

Statement of Originality

Contents

1. Introduction

2. Literature Review

3. Background on Canada’s Economy and Energy Industry

4. Data, Sources and Empirical Methodology

4.1 Data and Sources

4.2 Empirical Methodology

5. Empirical Results

5.1 Monthly Regression Results

5.2 Quarterly Regression Results

6. Conclusions

Bibliography

Data Appendix