Crazy Rain and the Little Dragon at Europe’s Door

(1)

Crazy Rain and the Little Dragon at Europe’s Door

A structural threshold vector autoregressive model for the Moroccan

business cycle

Erik Klok

1322753 June 25, 2012 Master’s Thesis Economics

University of Groningen

Department of Economics, Econometrics and Finance Supervisor: dr. J.P.A.M. Jacobs

Co-assessor: dr. G.H. Kuper

Abstract

Morocco has a highly volatile economy which hamper long-term development. This thesis develops a structural vector autoregressive model (SVAR) in order to identify and examine the sources of business cycle fluctuations in Morocco. Block exogeneity restrictions based on the small open economy assumption are imposed to capture the main characteristics of the Moroccan economy like international trade and financials linkages, and its exposure to erratic rainfall. Nonlinearities are allowed to account for threshold effects of rainfall during droughts.

The model reveals that rainfall is the dominant source of business cycle fluctuations in Morocco, even in the absence of a drought. These findings are in sharp contrast with the general believe that the Moroccan economy has become more resilient to rain shocks. International variables also have been a dominant source of business cycle fluctuations in Morocco, in particular trade price shocks. Although Moroccan authorities claim that the economy is resilient to international financial crises, the model shows otherwise. Migrant’s remittances are often praised for smoothing business cycle fluctuations, but they do not play a significant role in the Moroccan business cycle.

JEL classification: C32, Time-series models; C34, Switching Regression Models; C51, Model Construction and Estimation; E32, Business Cycle Fluctuations; E44, Financial Markets and the Macroeconomy; F24, Remittances; F41, Open Economy Macroeconomics; Q54, Climate, Natural Disasters, Global Warming.

(2)

Preface

Since the introduction of vector autoregressive (VAR) models by Christopher Sims in the eighties of the twentieth century – for which he was awarded the Nobel Prize in Economics in 2011-, they became a popular research area in applied macroeconomics. VARs are excellent frameworks for analyzing the transmission of shocks through an economy. The standard VAR model of Sims subsequently evolved over to time in order to cope with contemporary challenges like nonlinearities and the wish to include a relatively large set of variable while maintaining a parsimonious model.

In this thesis I apply the VAR framework to Morocco in order to identify and analyze the impact of various shocks to which Morocco is exposed. After a severe crisis in the late eighties, the Moroccan economy remained relatively volatile, reducing its long-term growth potential. It is therefore somewhat surprising that there exist no other attempts to identify and examine the sources of business cycle fluctuations in Morocco. A large VAR is constructed by taking advantage of the traditional small open economy assumption which results into block exogeneity restrictions on the model structure. Because of the large share of agriculture and limited available water resources, the Moroccan economy is vulnerable to rain shocks. Therefore, I was more or less obliged to include rainfall as a potential source of business cycle fluctuations in Morocco, which is a little bit controversial in the VAR literature. The constructed model incorporates threshold effects of rainfall in order to distinct the effect of rainfall on the economy under normal conditions for the effect during droughts.

Besides the modeling challenges, the construction of a decent dataset for Morocco was very demanding and several difficulties had to be encountered. For example,

meteorologist do not use political borders in their construction of a dataset, in contrast with economists. I therefore construct a rainfall index that is suitable for economic analysis. The construction of the dataset was a task in itself.

This thesis is organized as follows. In Chapter 1 an introduction to the construction of the VAR model for Morocco is given. Chapter 2 contains some background on the Moroccan economy, with a strong focus on climate in its relationship with economic activity. In Chapter 3 the VAR methodology is presented. Chapter 5 presents a description of the construction of the dataset. Chapter 6 provides the structure of the VAR model and a summary of the structural equations. Chapter 7 describes the computational model estimation and provides model diagnostics. In Chapter 8 impulse response functions and dynamic multipliers are provided. Chapter 9 provides an historical composition decomposition of the Moroccan business cycle. Chapter 10 contains a short summary, conclusions, and recommendations for future research and policy makers.

I would like to thank my supervisor dr. J.P.A.M. Jacobs for his endless patience with my inaccessible way of working and inappropriate communication skills. Furthermore, I would like to thank Mr. Younes Zouhar - a researcher at the IMF Middle East and Central Asia Department and former Chef de la Division de la Balance Morocaine des Paiments- who was so kind to provide me with data on remittances and tourism receipts.

Groningen,

(3)

List of Figures

4.1 Structure of the VAR model for Morocco 30

5.1 Graphical representation of the variables included in the SVAR model 68 7.1.A Foreign responses to foreign output, international financial stress and

foreign price shocks: foreign output, international financial stress,

foreign prices, foreign interest rates, import prices, and export prices 83 7.1.B Domestic responses to foreign output, international financial stress and

foreign price shocks: export volume, domestic expenditure, domestic

output, domestic prices, the exchange rate, and the domestic interest rate 84 7.1.C Responses to foreign output, international financial stress and foreign price

shocks: domestic financial conditions, tourism receipts, and remittances 85 7.2.A Responses to foreign interest rate, import price and export price shocks:

foreign output, international financial stress, foreign prices, foreign interest

rates, import prices, and export prices 86

7.2.B Responses to foreign interest rate, import price and export price shocks: export volume, domestic expenditure, domestic output, domestic prices,

the exchange rate, and the domestic interest rate 87 7.2.C Responses to foreign interest rate, import price and export price shocks:

domestic financial conditions, tourism receipts, and remittances 88 7.3.A Responses to export volume, domestic expenditure, and domestic output

shocks: export volume, domestic expenditure, domestic output, domestic

prices, the exchange rate, and the domestic interest rate 92 7.3.B Responses to export volume, domestic expenditure, and domestic output

shocks: domestic financial conditions, tourism receipts, and remittances 93 7.4.A Responses to domestic prices, exchange rate, and domestic interest rate

shocks: export volume, domestic expenditure, domestic output, domestic

prices, the exchange rate, and the domestic interest rate 94 7.4.B Responses to domestic prices, exchange rate, and domestic interest rate

shocks: domestic financial conditions, tourism receipts, and remittances 95 7.5.A Responses to domestic financial conditions, tourism receipts, and domestic

interest rate shocks: export volume, domestic expenditure, domestic output, domestic prices, the exchange rate, and the domestic interest rate 96 7.5.B Responses to domestic financial conditions, tourism receipts, and domestic

interest rate shocks: domestic financial conditions, tourism receipts, and

remittances 97

(6)

List of Tables

5.1 Individuals measures of financial conditions in Morocco 65 5.2 WMO stations and their appointed weights in the constructed rainfall index 67

5.3 List of variables in the SVAR model for Morocco 68

A2.1 Descriptive statics interpolation goodness of fit export and import series 126 A3.1 Correlations between individual financial components 127 A3.2 Principal component analysis: eigenvalues (components) 127 A3.3 Principal component analysis: eigenvectors (loadings) 127

A4.1 Rainfall regimes in Morocco 128

C1.1 System residual Portmanteau test for autocorrelation 130 C2.A System residual test: Cholesky (Lütkepohl) orthogonalization 130 C2.B System residual normality test: inverse square root of residual correlation

matrix (Doornik-Hansen) orthogonalization 131

C2.C System residual normality test: inverse square root of residual covariance

(7)

1 Introduction

After a near-bankruptcy in the early ninety eighties, Morocco successfully implemented a ten year structural adjustment program (SAP) of the IMF. Morocco became even known as ‘the little dragon at Europe’s door’ and was predicted a bright economic future. Unfortunately, the country could never deliver on its promise. One of the reasons for the disappointed growth rate since the ninety nineties is that the economy remained very volatile. Nevertheless, not attempts have been made so far to identify and examine the source of this volatility.

The purpose of this thesis is therefore to develop a model capable of identifying the major shocks to which the Moroccan economy is exposed since the ninety nineties and to understand the dynamic reactions these shocks generate. In addition this thesis measures the relative contribution of each of the shocks to the Moroccan business cycle over the last twenty years. Identification of the major shocks and their dynamic adjustment path is the first and essential step to make the Moroccan economy more resilient to shocks. For these

purposes, a structural threshold vector autoregressive model with exogenous variables and block exogeneity restrictions (STVARx) is developed for Morocco.

1.1 Morocco’s unfulfilled growth promise

A temporary boom and subsequent fall in the price of phosphate in combination with sharp rise in world interest rates, almost resulted in Morocco’s bankruptcy in 1983.1 A prolonged period of macroeconomic and financial instability followed. When revenues from phosphate exports collapsed –Morocco’s main export product at that time-, the ‘Moroccanization’ of the economy aimed at promoting agricultural self-sufficiency, discouraging imports and foreign investment, and state control over natural resources, came to an abrupt end. As a condition for new loans from the International Monetary Fund (IMF), the country agreed on a program to restructure its economy.2 The SAP resulted into a more diversified and trade oriented economy, and with country characteristics like its proximity to Europe, political instability, low wages and decent infrastructure, Morocco was predicted a bright economic future.

Morocco however never succeeded to fulfill this promise, mainly because of two reasons. Additional and necessary reforms that were planned for the early ninety nineties were delayed and proceeded far much slower than initially anticipated. In particular, progress on reforms aimed at reducing the role of the state in the economy were only slowly coming into place. Second, the Moroccan economy is still vulnerable to external and domestic shocks making output highly volatile. Because the economy has become more diversified due to the SAP program, it is likely to be more vulnerable to different sources of shocks. Before the reform program volatility arose mainly from two sources only: shocks to phosphate prices and bad harvests. Sekkat (2004) examines the Moroccan growth experience since the 1960 and distinguish three periods according to the behavior of the growth rate. The third sub-period (since the early ninety nineties) shows indeed much more volatile growth rates than the periods between 1960 and 1990.

The fact that high volatility of output hampers the long-term development of an economy has become widely accepted after the apex of the traditional macroeconomic dichotomy between volatility and growth. Lucas (1987) for example argues that the possible returns from understanding business cycle fluctuations are not worth the effort compared to

1_{The ratio of external debt to gross domestic output (GDP) reached 87, while the ratio of external debt to}

exports –Morocco’s main source of foreign exchange earnings at that time- reached an astonishing level of 305 (Reinhart, Rogoff and Savastano, 2003).

(8)

those from understanding economic growth. The identification and understanding of short-term economic fluctuations in Morocco is therefore of crucial importance for the economic future of the country.

1.2 Vulnerability to shocks

Since the reforms of the eighties, the Moroccan economy has become more diversified and less vulnerable to single shocks like fluctuations in the price of phosphate or harvests. At the same time when the economy became more diversified, Morocco gradually exposed itself to more potential shocks. For example, Morocco is today more exposed to international

financial markets and the tourism sector strongly developed. The increased international trade and financial linkages between Morocco and the world, particularly Europe, make the

country potentially more vulnerable to foreign shocks.

Nevertheless, one of the main risks Morocco continues to face is its exposure to a highly erratic rainfall pattern. Given the high dependency on rainfed agriculture, which typically accounts for some 15% of GDP but employs more than 40% of the workforce, private consumption and output are easily affected by changes in rainfall. In addition, Morocco is listed by the Food and Agriculture Organization (FAO) among the countries that face on average absolute water scarcity; water withdraws from renewable resources are no longer sufficient to satisfy all human demands. The total actual renewable water resources that are available in Morocco fluctuate with the level of rainfall. Rainfall therefore directly increase or decrease pressure on water resources and determines the level of water stress among mining, industry and other purposes. Because water sources are crucial for all socio-economic development, the effect of rainfall on socio-economic activity in Morocco is expected to be much larger than only the effect on rain-fed agriculture. Meanwhile, the rainfall pattern in Morocco is predicted to become even more volatile in the future. Rainfall deviations from normal levels during a year will increase both in magnitude and frequency.3 Or in words of the North- African people: ‘the rain goes crazy’.

Rainfall is often linked to the economy, with an extensive amount of literature focusing on the effects of long-term decreases in rainfall on economic growth (Barrios et al., 2003), the role of intra-annual rainfall variability in explaining growth differences between countries (Brown and Lall, 2006), or on estimating damage functions in order to estimate the effect of extreme rainfall events (e.g. Hallegatte et al., 2007). Meanwhile, in modern business cycle research rainfall has not been included as a potential source of short-run economic

fluctuations, despite the fact that many countries like Morocco are particularly vulnerable to relative small changes in rainfall levels during a year and not only to long-term decreases in rainfall trends. Interest in climate as a potential source of short term economic fluctuations in modern macroeconomics is gradually increasing since the work of Buckle et al. (2007) who successfully include a climate variable in their business cycle model for New Zealand.

Although not included in contemporary business cycle research before the work of Buckle et al. (2007), linking climate to economic cycles started already at the end of seventeenth century with the sunspot theory of Jevons (1886). Basically, Jevons (1886) linked the UK business cycle to the eleven-year sunspot cycle based on the relationship between sunspot activity and agricultural production. Although his theory was never taken too seriously, his thoughts on the transmission of shocks and anticipation effects are

3_{Fisher et al. (2002) designate the northwestern region of Africa as particularly vulnerable to changing rainfall}

(9)

interesting.4 Favorable harvests due to increases in sunshine reduce the prices of food and raw materials, thereby changing conditions in the money market. Manufacturing output increased because people have more resources left because food prices decreased. The effect of harvest fluctuations were multiplied by changes in investment decisions induced by self-fulfilling expectations of market participants responding to price signals.

The inclusion of a rainfall variable in the models for Morocco introduces some modeling challenges that emerge from the exogenous nature of rainfall and a potential non-linear relationship with economic variables. Threshold effects of rainfall in Morocco are expected based on Gober et al. (2001) and Baraket and Handoufe (1998) who show that rainfall sufficient to cause dramatic reductions in Morocco’s agricultural production is estimated at 38% below normal values, while below 19.8% there is no fear of any serious impact of rainfall on agricultural production. These threshold values are similar to Ameziane et al. (2007) who show that cumulative annual rainfall less or equal to 263 mm is less than normal water requirements for wheat crops and wheat production is altered drastic. Since the normal long-run average winter rainfall is estimated at 401.2 mm in this thesis, the deviation expressed in percentage points is 138 mm, or 34%. In additions slaughter rates increase more than proportional during droughts; Chaarani and Mahi (2007) for example show that the slaughter rate in Morocco increase to 16% in dry years, from 6% in humid years. A regime dependent binary variable is therefore introduced in the model to allow for some kind of non-linear behavior of rainfall during droughts.

Morocco’s highly erratic rainfall pattern is only one of the characteristics that makes the country vulnerable to shocks. Morocco is for example well known for its large share of diaspora across Europe. Migrant communities emerged since the 1960s from high demand for low-skilled labor in Europe. As the Moroccan people remain strong ties with their home countries, large flows of remittances followed. Remittances are in general considered as a welcome gift for a country because they provide funds for development and may smooth business cycle fluctuations. Nevertheless, remittances actually may amplify business fluctuations when the remitter is not altruistic, but has a portfolio-diversification motive instead. Under an altruistic motive remittances increase during recessions, while a portfolio motivates remitter will decrease its transfers in order to prevent losses. Remittances are also associated with Dutch Disease effects. The literature on the cyclical nature of remittances does not always adequately deal with the endogeneity problem that exists between the state of the Moroccan economy and the level of remittances. The vector autoregressive approach to the Moroccan business cycle as developed in this thesis is an appropriate framework to examine the cyclical nature of remittances without the danger of an endogeneity bias.

1.3 A VAR model of the Moroccan business cycle

The vector autoregression (VAR) model as developed by Sims (1980) has become a standard method in empirical macroeconomics to evaluate the dynamic response of an economy to shocks. Despite its popularity, a standard VAR model is not sufficient to capture the main characteristics of the Moroccan economy and is therefore insufficient for identifying the major shocks and their propagation mechanisms. Several important developments in VAR model specification and estimation since its introduction make it possible to construct a relatively large framework that captures the most important characteristics of the Moroccan

4_{Until today, no empirical evidence has been found that there exist a causal relationship between sunspots and}

(10)

economy. The VAR model that is constructed for Morocco uses the innovative block exogeneity procedure of Cushman and Zha (1997) and Zha (1999) that makes use of the small open economy (SOE) assumption in order to make it possible to include a relatively large number of variables compared to a standard VAR. A block structure reduces the numbers of parameters that must be solved for at all times. At the same time, a block structure brings some actionable clarity to the complex dynamic relationships and reveals some of the deeper dynamics of a VAR. Block exogeneity procedures are imposed for example by Dungey and Pagan (2000) in order to construct a VAR model for Australia. Buckle et al. (2007) build on the work of Dungey and Pagan by extending a small-open economy VAR model for New Zealand with a climate block in additional to an domestic economy, foreign economy and international trading prices block.

This thesis develops a structural vector autoregressive (SVAR) model to identify and examine the main sources of Moroccan business cycle fluctuations and follows the block structure procedure of Buckle et al. (2007) that arises naturally from the assumption that Morocco is a small open economy subject to external and internal shocks. A key innovation of the VAR model for Morocco is the incorporation of nonlinear effects, in the spirit of Chan et al. (1995). Chapter 4 provides an extensive overview of the model developed in this thesis and the distinctive character compared with previous similar models. Impulse response functions and dynamic multipliers are used to evaluate the dynamic impact of the various shocks that are included in the model. Subsequently, a historical decomposition is performed to estimate the contribution of each individual shock to the Moroccan business cycle over the last twenty years.

Several interesting results emerge from the developed VAR model for Morocco. Despite its developed towards a more sophisticated economy, rainfall remains the dominant source of business cycle fluctuations in Morocco, even in the absence of droughts. This suggest that the Moroccan economy has no become more resilient to rain shocks. As expected, due to

reforms, the financial sector has become more important, but impulse response functions suggest that financial reforms should continue. Financial conditions seem to adjust only at a very slow phase when the Moroccan economy is hit by a shock, reducing the recovery speed of the economy. Reforms aimed at increasing the mutual confidence between financial institutions and economic agents should reduce the long-lasting effects of shocks on financial conditions in Morocco. International variables also have been a dominant source of business cycle fluctuations in Morocco, in particular trade price shocks. Despite the claim by

Moroccan authorities that the economy is resilient to international financial crises, stress on international financial stress directly affects the Moroccan economy. Migrant’s remittances are often praised for smoothing business cycle fluctuations, but they do not play a significant role in the Moroccan business cycle.

Finally, puzzles that are often find in impulse response functions based on VAR models like the liquidity, price, exchange rate, forward discount bias and interest-ouptut puzzles (see Kim and Roubini, 2000 and King and Watson, 1996) do not emerge in this thesis, which suggest an appropriate identification scheme of the model. The inclusion of the forward looking indicators like rainfall and international financial stress may have

(11)

2 Background

This chapter provided some background information on the transmission mechanism of rainfall on the Moroccan economy and migrant’s remittances. Background information on other aspects of the Moroccan economy like the monetary and exchange rate policy, trade patterns, is incorporated in Chapter 4 that provided a description of the structural equations of the model.

2.1 The economic impact of rainfall in Morocco

The Moroccan economy is not only very vulnerable to fluctuations in rainfall because of its large share of rainfed cereal agriculture, but also because its low balance of total renewable water resources which effect almost all economic activity in the country. Before discussing the basic impact of climate on the economy, a brief overview of the Moroccan climate is provided.

2.2.1 The climate in Morocco

Morocco is located at the northwestern corner of Africa and is surrounded by the North Atlantic Ocean, the Mediterranean Sea, the Atlas Mountains and the Sahara, which makes the country’s climate very different across regions as well as across seasons. There exist a Mediterranean or a dry-summer subtropical climate in the northern coastal regions that is characterized by warm summers and mild and wet winters. The southern interior regions of Morocco have high volatile temperatures and are often hit by droughts. Because of these unfavorable climate conditions in the south of Morocco and in the mountains, economic activity mainly takes places in the coastal areas and their hinterland where climate is more favorable for humans. This is basically the reason why the rainfall dataset that is used in this thesis need to be adjusted to account for the regional differences. The construction of the rainfall index is elaborated in Chapter 5.

Temperatures in regions with economic activity are generally moderate with a small range between winter low and summer high temperatures, while rainfall patterns are much more volatile during the year. Temperature is a rather conservative element of the Moroccan climate with a relatively small annual range, while rainfall is the most defining characteristic (Oliver, 2005). Subtropical high pressure cells during summer makes rainfall scarce, while during winter the polar jet stream brings abundant rain. Although the variation in total winter rainfall is relatively low, monthly variation during the wet season during November-March, is very erratic. This volatile rainfall pattern is directly related to the North Atlantic Oscillation (NOA) and the El Niño-Southern Oscillation (ENSO). NOA is a regular climatic

phenomenon in the North Atlantic Ocean of fluctuations in the difference of atmospheric pressure at sea level and controls the strength and direction of westerly winds across the North Atlantic. The magnitude of these westerly winds determine the level of winter rainfalls in Morocco; when winds are strong rainfall is heavy. This rainfall pattern created by the NOA is however on average - but over a period that varies from three to seven years - every five years disturbed by the ENSO. One variant of ENSO is El Niño which is a prolonged warming over the east-central tropical Pacific Ocean and brings in more rain than on average, but is magnitude is varying, from almost zero to extreme cases as in September 2008 when record rainfall levels were denoted. La Niña in contrast, the name for the cold phase of ENSO, coincide with less than average rainfall levels.

(12)

replenishment of the available renewable water resources. Water is crucial for all kinds of economic activity. Because water balance are generally low in Morocco,

relatively modest changes in rainfall levels may change the level of competition for scarce water resources between agricultural, industry and domestic use. Those effects of rainfall on the economy is like to initiate a dynamic reaction that is described in Chapter 7 by hand of dynamic multipliers resulting from a rain shock. Both channels are explained in this paragraph.

2.2.2 Rainfall and agricultural production

Almost 90% of agriculture in Morocco consist of rainfed cereal production and although the share of agriculture in GDP is with 18% relatively modest, the sector employs more than 40% of the labor force and is therefore of crucial importance for the Moroccan economy.5 Changes in cereal production levels result in not only rural household behavior, but also urban

household behavior. The government of Morocco is also seriously affected by changes in cereal production due to imposed price controls and changes in tariff revenues due to changes in required cereal imports.

The agricultural sector can be divided into a rainfed traditional sector consisting of cereal (wheat, barley and maize) production for domestic consumption and a relatively small irrigated and more modern export oriented sector producing fruits and vegetables. Due to a lack of irrigation systems in the traditional sector in combination with limited soil-water storage capacity, the major share of agriculture in Morocco is exclusively directly depended upon adequate levels of rainfall.6 Cereal production has become particularly vulnerable as a consequence of price controls imposed by the Moroccan government in the late 1980s which resulted in artificially high prices for cereals. As a consequence, cereal production expanded towards unproductive areas with low and volatile rainfall levels. In addition, farmers

expanded wheat production at the expense of the less rainfall dependent crops like barley. As a results, the standard deviation of agriculture growth in Morocco increased from 10 percent in the 1970s up to 42 percent in the 1990s as illustrated by Azzam and Sekkat (2005). The same authors also show that total-factor productivity in Moroccan cereal sector in the long and short run is driven by changes in rainfall and technical efficiency and not by government policy, demand effects and technical changes. The effect of rainfall on cereal production is also empirical examined by Heng et al. (2007) who show that in the case of wheat, yields in Morocco during the past twenty years were frequently limited by the amount of - and timing of rainfall. Growing season precipitation amount is the best single predictor of above net primary productivity of grassland or crop production (see for example Nippert et al., 2006 and Cooper et al., 2008). Barakat and Handoufe (1998) define threshold levels for rainfall that are sufficient to cause a significant reduction in agricultural production in Morocco. The security threshold is estimated at 19.8% below normal rainfall during October and May, and indicates a rainfall deficit without fear of a decrease in agricultural production. A significant decrease in agricultural productivity is certain when rainfall is 38.1% lower than average. More information on this kind of threshold effects are given in Chapter 5. Not only the amount of rainfall is important, but also its timing. The cereal growth cycle in Morocco coincides with the rain season during October to May, with seeding occurring typically in October-December, tillering and stem extension during January-February and heading and ripening in March-May.

Gober et al. (2001) indicate that cereal yields in Morocco are linearly related to cumulative rainfall in January through March. This may be explained by the fact that despite

5_{Source: UN Statistics Division - National Account Main Aggregate Database.}

6_{Droogers et al. (2001) calculate a worldwide soil-water storage capacity (SWSC) index that shows the}

(13)

the typical growing cycle, famers could theoretically wait for the rain up to March making it difficult to readjust production plans in the case of low rainfall before March although this is not desirable (Karaky, 2002).

2.2.3 Household behavior in response to changes in agricultural production

As a consequence of vulnerability of crops to rainfall, farmers in general tend to respond immediately to changes in rainfall by reducing investment or even stop cropping and focus on livestock management in response to dry spells. Famers cannot postpone seeding, which makes it making it difficult to readjust production plans in the case of low rainfall before March (Karaky, 2002). This point is also indirectly reflected by studies that found that cereal yields are linearly related to cumulative rainfall in January through March (Gober et al. 2001). The response of livestock capital in Morocco to rainfall conditions on livestock capital seems to be in line with other countries with limited access to financial services. Livestock is not only threatened by the absence of drinking water as a result of decreases in rainfall, but several studies (Karaky, 2003 and Lybbert et al., 2009) find that livestock capital in Morocco responds accordingly to changes in cereal production levels. Livestock act therefore as an asset for farmers to hedge against rainfall fluctuations, although farmers often complain that they are forced to sell livestock at difficult times when prices are low.

Lybbert et al. (2009) emphasize the potential limited possibilities in Morocco regarding the mitigation of income deficits by reallocating assets. Due to the possible existence of a critical herd size threshold in the case of livestock, below this value the marginal productivity of livestock falls sharply, the cost of current consumption in terms of future productivity increases. This is reinforced by the fact that for credit-constrained rural households in Morocco, productive assets, in particular livestock, can be difficult to recoup. Hence, household are expected to sell livestock only when it is strictly necessary. In a survey of 250 Moroccan rural households Lybbert et al. (2007) indeed show that households will first draw down stores of agricultural products if possible and, borrow money from family or friends, followed by the selling of livestock and renting out land. There is however a big difference in types of livestock: sheep and goats are rapidly sold after changes in rainfall, while selling of cattle is held as a last resort indicating that a critical herd size exist for cattle and they are difficult to recoup, in contrast of sheep and goats which are not considered to increase to cost of current consumption in terms of future productivity. Buckle et al. (2007) also refer to the existence of a positive relationship between livestock capital and weather conditions in New Zealand as a possible explanation of the observations that export volumes initially increase in response to adverse weather conditions due to a possible increase in slaughter rates.

Secondary effects of changes in cereal production levels in response to rainfall are examined more formally by Karaky (2002) in a general equilibrium model that is based on a social accounting matrix. The model reflects the interaction between production, income,

consumption and capital accumulation of Moroccan households, companies, the government and the rest of the world based on 1997 national income values. One of the interesting features is that Karaky makes a distinction between six different types of households in Morocco, that respond differently to changes in cereal production levels. Households welfare is affected by changes in cereal production levels by changes in returns to factors of

(14)

change in purchasing power is much larger for rural and urban poor. Their purchasing power available for food is directly affected by changes in cereal production levels, while this is not the case for medium and large farmers that are net sellers of food. Urban non-poor

households suffer therefore less than poor household although they also face higher food prices, but they are partly compensated by an increase in returns to urban capital as

production activity would tend to shift towards urban commodities when cereal production declines.

2.2.4 Fiscal and monetary responses to changes in agricultural production

Changes in cereal production has far more widespread implications than only affecting the income households. Because Morocco is structurally a net importer of cereals, favorable rainfall dramatically declines cereal imports. To maintain the fixed exchange rate, net capital inflows decrease to balance the current account. In addition, tariff revenues fall in line with the decrease in imports and government savings decrease as consumption is assumed to be remained fixed in the short-term. As a consequence of the fall in foreign savings and government savings domestic investment will decrease. The opposite occurs with

unfavorable cereal production levels. Finally, Globe et al. 2001 mention that loan default by farmers during adverse climate conditions imposes additional cost on the Moroccan

government because of its program of debt relief during times of rainfall deficits.

In addition to the effects of rainfall on cereal production and livestock capital, it also affect pasture growth and everything associated like lambing percentages, cattle slaughter rates and dairy production. And more importantly, the effect of changes in agricultural production in Morocco is particularly large as agriculture is also an important source of intermediate inputs for several manufacturing industries in Morocco, like the flour and pasta, and leather processing industries.

As becomes clear from the work of Karaky (2002), rainfall effects on agriculture in Morocco are likely to spill over to non-agricultural activity like returns to urban capital or fiscal and policy consequences. In addition, there exists in the literature a long history of the the independency of agricultural and monetary factors causing business cycles. In this view, elaborated by for example Andre (1906) movements in agricultural production alters money demand and the associated change in interest rates induce an economy-wide impact. Due to the existence of such a relationship, monetary policy might be effective in mitigating the effect of a rain shock on the economy. However, Morocco has limited scope to pursue an independent monetary policy given its pegged exchange rate regime, as will be discussed in Chapter 5.

2.2.5 Rainfall and total actual available renewable water resources

In addition to the relatively large share of rain-fed agriculture that directly dependents on levels of rainfall, rainfall affects the Moroccan economy indirectly by altering scarce water resources needed for industrial, mining, irrigated agriculture and other human demands. Rainfall also drastically changes government expenses due these scare water resources. For example, expenses for programs to protect livestock, forest and adequate water supply for villages as a consequence of the rainfall deficit in the period October 1999- March 2000, were estimated at 6.5 million Moroccan dirham (Gober et al., 2001).

The availability of water from rainfall is limited by the relatively high rates of

evapotranspiration in Morocco.7 These water resources consist mainly of so called blue water that originates from rainfall that escapes evapotranspiration and is captured in rivers and

7_{The amount of water that remains in the soil after what is evaporated and what is transpired by plants as a part}

(15)

aquifers. The other source of freshwater availability, green water, that is formed by infiltrated rain and is of less importance for Morocco due to low soil storage capacity. This also

explains the large share of rainfed agriculture. During the past twenty years, Morocco is increasingly suffering from a constant water deficit caused by irregular climate conditions in combination with increased demand from water due to increased economic activity.

Total internal renewable water resources in Morocco are theoretically on average estimated at 29 billion m3 per year8 -although the technically and economically feasible resources are much lower -and with a population of 32.3 million9, Morocco is with less than 1000 m3 per capita per year defined by the Falkenmark Water Stress Indicator (Falkenmark et al., 1989; 2007)as a country that faces physically water scarcity. Water scarcity can lead to water supply collapse, crop failure in irrigated fields, the closure of river basins and increased infrastructure cost to make more water accessible for economic use. For example, Yang et al. (2003) conclude that there is a threshold of about 1500 m3/capita per year below which a country’s cereal imports become strongly inversely correlated with its renewable resources. In addition, the total annual withdrawals from these low levels of water resources exceeds 40% in Morocco which is in the literature a threshold value to indicate high water stress (Raskin et al., 1997 and Alcamo et al., 2000). Water is considered as a constraint on socio-economic development and water supplies are no longer adequate to satisfy all human demands.

Because internal renewable water resources include the average annual flow of rivers and the recharge groundwater generated from rainfall, even slight variations in these on average low water resources induced by changes in rainfall may affect the economy. Rainfall in Morocco therefore directly tightens or weakens the constraint on economic activity imposed by scarce water resources. Although agriculture (mainly irrigated exports) still accounts for 87% of the total amount of water withdraw, municipal (9.7%) and industrial water withdraw (2.9%) are gradually increasing pressure on water resources. Two main water users, phosphate mining and hydro electricity generation are worth mentioning because of their special roles in the Moroccan economy.

The cause of the economic crisis during the 1970s and 1980s were fluctuations in world phosphate prices and although Morocco has become less dependent on phosphate mining since the IMF SAP, phosphate and its derivatives still account on average for 18%of total Moroccan export value.10 Although Morocco is the world’s third largest miner and processor of phosphates and first exporter housing approximately two third of the world’s phosphate reserves, the phosphate industry activity is constrained by scarce water resources in

Morocco.11 Phosphate mining requires massive quantities of water: according to the Florida Institute of Phosphate Research, a single phosphate mine pumps on average more than 100,000 gallons of water a minute, and can be evaluated to 57 million m3, or in terms per capita 2 m3, in Morocco. With already scarce water resources, these huge water demands of the phosphate mines and their chemical branches impose heavy burdens on other

socioeconomic sectors in surrounding areas like agriculture and the drinking water industry. Phosphate mines increase the pressure on already scarce water resources, and rainfall becomes even more crucial in areas surrounding phosphate activity.

8_{Source: FAO AQUASTAT.}

9_{Source: Population Division of the Department of Economic and Social Affairs of the United Nations}

Secretariat, World Population Prospects: The 2008 Revision.

(16)

Furthermore, the Moroccan Ministry of Energy aims to expands its hydropower generation from 8% - but even 25% of total energy supply is currently of hydraulic origin - to 20% in 2020. Hydropower production variability is proportionally related to the precipitation level by altering river flows. River basins have a tendency to amplify precipitation changes by

approximately factor 2.6 (Harrison et al. 2006), e.g. runoff changes tend to be greater than the precipitation change causing them. Adequate water balances in basins are needed to

guarantee the fall and movement of water during period of insufficient rainfall. Water levels in basins during the year depend on rainfall levels during winter times. When rainfall is insufficient during winter, low water balances in the summer hampers hydropower generation in that period. With sufficient water balance during a year it does not matter if rainfall patters are volatile as long as the total level of rainfall is adequate to fill the water basins to guarantee an equal stream of water is guaranteed during the year. With low water balance as is

generally the case in Morocco, a moderate rain shock may already affect the generation of hydropower resulting in fluctuations of energy prices.

To summarize, the Moroccan economy is not only affected by rainfall due to agricultural production, but also by the competition amongst socioeconomic demands for water resources due to low water balances. For example phosphate mines reduces water resources left for irrigated agriculture heavily, making them quite vulnerable to small changes in rainfall that is the indirect source of irrigation. In the future, pressure on water resources will become even larger in line with Morocco’s predicted economic development, making the economic activity even more vulnerable to fluctuations in rainfall.12 Both the direct and indirect impact of rainfall on the economy will trigger a dynamic multiplier effect by altering economic variables like prices and savings. This dynamic path is analyzed in Section 6.

2.2 Migrant’s remittances

Morocco is well known for its large diaspora communities across Europe, that started to settle during the 1960s in response to Europe’s high needs for low-skilled labor. The general belief at that time was that Moroccan guest workers would accelerate the development of their home country’s economy by temporary relieving pressure on the unbalanced Moroccan labor market, providing funds for investment and redistributing knowledge when they

returned.13As it turned out, most workers did not return and migration continues until today with already 4 million people of Moroccan decent live abroad, while Morocco has

approximately 33 million inhabitants.14

Because Moroccan migrants in general remain close ties with their relatives at home, large remittance flows arose and gradually became an important element of Morocco’s economy. Workers’ remittances currently contribute 7% to GDP -which corresponds with the recent twenty years average- and Morocco is the world’ fourth receiver in absolute terms.15

Remittances are expected to remain substantial in the near future as migration will continue for at least the coming two decades as predicted de Haas and Plug (2006). Their economic impact remains ambiguous, despite that remittances in general are considered as a welcome gift to an economy.

Remittances are in the literature associated with reducing poverty rates (Adam and Page, 2010), increasing investment in education and health care and promote development of

12_{See for example Taleb (2007) for future scenarios on water demand in Morocco.}

13_{See de Haas (2005) and Leichtman (2002) for both political and economic history of Moroccan migration.} 14_{de Haas (2005) and UN World Population Prospects.}

(17)

financial institutions (Aggerwal et al., 2006) which in its turn fosters growth (Mohamed and Sidiropoulos, 2010), overcoming problems of credit rationing (Edwards and Ureta, 2003) and Yang, 2007), and promoting entrepreneurship (Maimbo and Ratha, 2005).16 There are also several potential negative externalities of remittances on the micro-level like

moral hazard problems, crowding out the insurance aspect of wages and thereby sub-optimally shifting risk from risk neutral firms to risk-adverse remitters resulting into higher wage volatility over time and lower work effort (Chami and Fischer 2000), fueling

speculative bubbles in unproductive real estate, increasing income inequality, encouraging migration (Van Dalen, Groenewald and Fokkema 2005) and thereby increasing brain drain (Leichtman 2002).

While these are all effects of remittances on the micro level, remittances also have a serious impact on the macro level directly. Remittances act as stable source of foreign exchange which reduce the probability of sudden stops and current account reversals (Bugamelli and Paterno, 2009), provide a collateral for bond issuance which lowers the cost of debt (Ketkar and Ratha, 2010), reduce the volatility of output by smoothing consumption in economic downturns, and may even foster growth (Mohamed and Sidiropoulos, 2010; Berthomieu and Tykhonenko , 2009). Unfortunately, remittances also impose a number of serious threats to an economy that are relatively under exposed, like generating Dutch Disease effects (Amuedo-Dorantes and Pozo, 2004) or amplifying business cycle

fluctuations. Negative externalities might even surpass the benefits of remittances, leading some studies to reject the conventional wisdom that remittances are good for growth (Glytsos, 2005; Chami et al. 2003, 2008).

Most of those studies are however based on cross-country samples, rather than individual countries despite the behavior and effect of remittances depend on individual country characteristics. For Morocco only a few country- specific studies are available. In addition several existing studies suffer from a potential endogeneity problem when examine the effect of remittances on economic growth. Most studies focus on the relationship running from remittance to growth; while a reverse causation is also possible. Economic growth in the receiving country determines the level of remittances. Two distinct hypotheses exist on the behavior of remittances in response to economic conditions in the receiving country, and depend on the motivation of the remitter. The most accepted hypothesis is that the remitter is altruistic: he compensates family members in his country of birth during times of hardship. In this case, remittances respond negatively to changes in income of the receiver (Chami et al., 2008; Yang and Choi, 2007). The other view is based on a portfolio-diversification or rent-seeking motive of the remitter and remittances responds positively to changes in income of the receiver (Amuedo-Dorantes and Pozo, 2010).

For the Middle East and North Africa (MENA) region, Glytsos (2005) finds an overall positive long-run relationship between remittances and growth in Morocco. Berthomieu and Tykhonenko (2009) show by hand of a panel convergence framework that for six MENA countries –including Morocco- there exist a positive impact of remittances on growth. Mohamed and Sidiropoulos (2010) also conclude for the MENA region by hand of fixed-effects panel methods that remittances have a positive impact on growth. Interestingly, Sayan (2006) shows that in the short-run remittances in Morocco tend to be procyclical.

Remittances amplify business cycle movements, rather than smoothing them.

16_{Docquier and Rapoport (2005) provide an extensive literature overview on the microeconomic impact of}

(18)

3 Methodology

Contemporary business cycle research is dominated by two methodologies: structural vector autoregressive (SVAR) and dynamic stochastic general equilibrium (DSGE) models. Both models developed in response to shortcomings of the Cowles Commission structural equation approach (SEMs), which was the main paradigm of empirical macroeconomic research in the 1950s, 1960s and 1970s. Macroeconomist became divided after Lucas’s critique (Lucas, 1976) that questioned the constancy of parameters in SEMs. Rational expectations of forward looking agents result in misspecified SEM models under Keynesian identification

assumptions because structural parameters are not invariant to changes in policy rules. Furthermore, identification of SEMs was mainly achieved by imposing restrictions that take the form of exogenous variables that appear in one equation but not in another. The problem with this way if identification is that true exogenous variables are hard to find.

Macroeconomists subsequently moved in two directions; one group favoring dynamic stochastic equilibrium models (DSGE), while the other group beliefs in the vector

autoregressive (VAR) approach to macroeconometric analysis as introduced by Sims (1980) for which he was awarded the 2011 Nobel Prize in Economics. Today, models tend to incorporate elements of both approaches.

3.1 Dynamic stochastic general equilibrium models

The first group of economists continued to follow the Cowles methodology, albeit

introducing the assumption that data is generated by a dynamic general equilibrium model in which forward looking agents have rational expectations of future prices. These models evolved into sophisticated dynamic stochastic general equilibrium (DSGE) models reflecting New Keynesian and real business cycle (RBC) paradigms. DSGE models based on the New Keynesian paradigm attempt to provide microfoundations for Keynesian concepts such as nominal price rigidities and non-neutrality of money. In this spirit, Rotemberg and Woodford (1997) were the first to introduce an econometric specification from an explicit model of intertemporal optimization on part of the suppliers and the purchasers of goods and services to evaluate monetary policy rules.

(19)

the economy works, but they cannot fulfill the requirements to match in detail the dynamic behavior of accounting constructs and proxy variables that make up the data. At best, DSGE can be used a mechanism for generating a prior (for a VAR), not a model of the data like VARs. See Tovar (2008) for a more comprehensive discussion on the applications and limitations of DSGE models.

Potential misspecification of DSGE models result from using Bayesian techniques with incorrect priors. These DSGE models give less satisfying answers to policy questions than an alternative less perfect fitted model when employed in the estimation of a dynamic shock process. Calibration of DSGE models may therefore be a more successful approach (Kocherlakota, 2007). For Morocco, calibration of an extensive DSGE model is not an attractive option because micro evidence on optimal household and firm behavior is generally unavailable, in particular for large sample periods. Furthermore, the critique on Bayesian estimation of DSGE models and the general concerns on the degree of misspecification, make the construction of such a model for Morocco not attractive.

3.2 Vector autoregressive models

The second group of economists that became dominant after the Cowles Commission uses vector autoregressive (VAR) models that were developed by Sims (1980) in response to the shortcomings of the traditional identification of SEM models. According to Sims those restrictions on the short-run dynamics of the model are ‘incredible’. Because true exogenous variables that are required for identification in SEM models are hard to find, Sims proposed to treat all variables in the model symmetrically. He developed an a-theoretical unrestricted system of reduced form equations where each variable depends on its own lags and the lags of all other variables included. This system is known as a standard our educed form vector autoregression (VAR) and treats all variables as endogenous. The data is therefore free to estimate the relationships between the variables without any restrictions.

However, because VARs are reduced form models, some structural restrictions are required to identify the relevant innovations and subsequently impulse responses. Some form of identification is needed because reduced-form errors are generally correlated. Sims (1980) proposed to use a type of recursive system where a Choleski decomposition is used to achieve identification; contemporaneous shocks to higher ordered variables feed through to lower ordered variables but not vice versa. Unless the underlying structural model can be identified from the reduced-form VAR, the innovations in a Choleski decomposition have no economic interpretation. In the end, the model of Sims is subject to the same critique Sims elaborated regarding the identification approach of the Cowles Commission.

In response to the short-coming of recursive VARs that use a Choleski decomposition, Sims (1986) and Bernanke (1986) proposed the use of economic theory to achieve an appropriate identification, resulting into so-called structural vector autoregressions (SVAR). A SVAR uses assumptions based on economic theory to sort out the contemporaneous links among the included variables. Stock and Watson (2001) define SVAR models as reduced form VARs that require identifying assumptions that allow correlations to be interpreted causally. This produces instrumental variables that permit the contemporaneous links to be estimated using instrument variables. For example Bernanke (1986), Blanchard and Watson (1986), and Sims (1986) recover the structural innovations from the residuals by using economic theory rather than an arbitrary Cholesky ordering of the residuals.

(20)

causes and therefore it is natural to treat them as approximately uncorrelated. SVAR models therefore examine the dynamics of a model by subjecting it to an unexpected shock. SEMs do not make a distinction between expected and unexpected changes in one of the included variables. Structural innovations are in SEMs interpreted as errors that reflect the effects of omitted, nonessential variables and are unrestricted. Today, VAR models are widely used in macroeconomics; for example they have been used to examine the effects of monetary policy on output (Sims, 1986), to distinct the relative importance of supply and demand shocks on economic fluctuations (Blanchard and Quah, 1989), to decompose real and nominal exchange rate movements into the components induced by real and nominal factors (Enders and Lee, 1997), and more recently to document the effect of climate on the business cycle (Buckle et al., 2007).

3.3 Cointegration: vector error correction models and global VARs

SVAR business cycle models traditionally focus on short-run fluctuations around a long-term trend, where the long-term trend is filtered out in order to work with stationary data.

However, it is possible to incorporate nonstationary variables in the model, when variables are cointegrated. Equilibrium relationships assure that variables cannot move independently of each other, which ensures that the dynamic path of such variables must bear some relation to a current deviation from the equilibrium relationship. An ordinary SVAR fails to display such connection between the change in a variable and the deviation from equilibrium as expressed in an error correction. The short-term dynamics of the variables in the system are influenced by deviations from equilibrium. Pesaran et al. (2004) use the apparent

synchronization of international business cycles to construct a global vector autoregressive (GVAR) model. Cointegrating relationships allow for interaction amongst different economies and account for the global comovement between various macroeconomic variables like output, inflation, interest rates, equity prices, oil prices, money supply and exchange rates. Country specific vector error-correction models are combined to generate forecasts for all the variables in the world economy simultaneously.

In a similar way, Dungey and Pagan (2009) extend their original structural VAR model for Australia (Dungey and Pagan, 2000) by allowing for both transitory and permanent components in the time series, in contrast with the previous framework that models

macroeconomic outcomes as transitionary deviations from a deterministic trend. The resulting structural vector error correction model (SVECM) has a theoretical orientation but also reacts to the nature of the data, specially the evidence for co-integration. Although the inclusion of long-term behavior in a business cycle model is appealing, it becomes very complicated when the number of variable that are included in the model increase. Multiple cointegrating relationships are hard to explain from economic theory and assumptions required to identify the multiple cointegrating relationships are often difficult to justify. The approach of Dungey and Pagan (2009) is to treat the terms of trade, interest rates and prices as variables integrated of order zero [I(0)] after eliminating a deterministic component. Meanwhile foreign output, exports, GNE and GDP are defined as variables integrated of order one [I(1)] and are included in a test for the presence of cointegration between the variables. In these tests, exports and foreign output are treated as exogenous, well the cointegration test is augmented for the I(0) variables – the terms of trade, interest rates and prices. A major drawback of cointegration analysis is its extreme sensitivity to tests

(21)

interest differential. This problem is elaborated by Kennedy (2003) who argues that when testing for cointegration, if a relevant variable has been omitted the test should fail to find cointegration. In the context of Dungey and Pagan (2009) this may be a problem as they exclude stock market capitalization in their 2009 compared to its 2000 model, however, the trends of GDP and stock market capitalization are likely to be cointegrated.

With their test specifications and model set-up, Dungey and Pagan found evidence for two cointegrating vectors. In this case when the cointegration rank exceeds one, it is not straightforward to interpret the cointegrating vector because any linear combination of these cointegrating vectors is also a cointegrating vector. Therefore the separate behavioral relationships need to be identified by appropriately restricting the individual cointegrating vectors. This is a major challenge associated with cointegration analysis. Greenslade et al. (2002) show that statistical techniques proposed in the literature for identification in cointegrated systems are almost impossible to implement successfully.17 Therefore they recommend a thorough use of economic theory at an early phase by imposing exogeneity restrictions and restricting the dynamic adjustment of the model to increase the power of tests of overidentifying restrictions on the long run cointegrating vectors. Dungey and Pagan (2009) identify the two cointegrating relationships by imposing zero restrictions that are examined by a chi-squared test for binding restrictions.

Kennedy (2003) however also emphasizes that it is important that arguments form the basis for imposing identification restrictions because of the difficulty of interpreting estimates of the cointegrating vectors. If economic theory tells us that tow variables should be

cointegrated (for example relative inflation rates and the exchange rate in the case of

purchasing power parity theory), then testing for cointegration is used as a test of that theory. In a model with a relatively large set of variables, it is likely that multiple cointegrating relationships arise that are unexplainable. It is therefore not attractive to incorporate

cointegrating relationships in a business cycle model with a relatively large set of variables. Meanwhile, the inclusion of fewer variables increases the misspecification of cointegration relationships due to omitted variables problems. The results of Dungey and Pagan (2009) however are quite similar to their earlier work, except from the impact of monetary policy shocks, which seems to be slightly overstated in the old model. However, the models are not completely comparable because of a different set of variables of the unadjusted 2000 model.

3.4 Hybrid DSGE-VAR models

Some contemporary research combines both the VAR and DSGE approaches resulting into hybrid DSGE-VAR frameworks, as is done for example by Del Negro and Schorfheide (2006) and Lees et al.(2010). This modelling approach however received heavy criticism. Favero (2007) for example argues that the logic of criticism moved to the Cowles

Commission approach of the 1970s might apply exactly to DSGE-VAR models. In fact, Favero claims that the DSGE-VAR approach is looser than the Cowles Commission

methodology in that restrictions are not imposed and tested but restrictions are made fuzzy by imposing distributions on them. Hence, an arbitrary amount of uncertainty is added to model based restrictions in order to make them compatible not with the data but with a model-derived unrestricted VAR representation of the data.

17_{Like the widely used Johansen (1988, 1991) procedure that forces all cointegrating relationships to be}

(22)

3.5 Business cycle framework for Morocco: the SVAR methodology

Based on the previous discussion on DSGE, SVAR , DSGE-VAR and SVEM models, the SVAR framework is chosen to be the foundation of the business cycle model for Morocco. Advanced techniques for identification are used to overcome traditional shortcomings of VARs like the relative small number of variables that can be included because degrees of freedom are quickly consumed as the number of variables or lags

increases. Nonlinear behavior resulting from climate is also incorporated. The specific design of the SVAR model for Morocco is elaborated in Section 4 below.

Although most economists agree that DSGE models are important for understanding economic behavior, they at least require further development regarding the incorporation of price stickiness and financial market frictions (Kocherlakota, 2010). This requires even more data and understanding of microeconomic agents, which is for a less developed country like Morocco unavailable. Moreover, the main critique on DSGE is how they deal –or neglect- market imperfections, and as I elaborated in previous sections of this thesis, the Moroccan economy is characterized by many market imperfections like labor market rigidities, corruption and credit constraints. Although markets are less complex than in developed countries, the many market imperfections and lack of behavioral data and understanding limit the success of the construction of a DSGE model for Morocco. For example, we do not know how agents in Morocco respond to the receipt of remittances, do they increase consumption or investment? What is the motive of the remitter? Are they altruistic motivated and compensate loss of income at home or are they rent-seeking and increase investment during booms? How do Moroccans respond to changes in rainfall? Do they for example anticipated by reducing their livestock during droughts? We do not know. Since DSGE models start from microeconomic principles of constrained decision-making, instead of given observed

correlations, they are difficult to construct for Morocco. To return to the critique of Sims on DSGE models, making assumption on microeconomic behavior in Morocco when

construction a DSGE model for Morocco likely cannot fulfill the requirements to match in detail the dynamic behavior of accounting constructs and proxy variables that make up the data. A VAR model therefore likely fits better than a DSGE model when applied to real Moroccan data. Incorporating cointegrating relationships is also not an attractive option as discussed above when the number of variables included in the model is relatively large. To understand the VAR methodology, consider the k-vector times series generated by the k-dimensional structural vector autoregressive process with exogenous variables VARX(p), ignoring a constant term for simplicity of exposition.

Γ() = Γ()+ (3.1)

where is vector of endogenous variables, Γ() = − ∑ denotes polynomials in the lag operator of the endogenous variables, is a vector of exogenous variables, Γ= + ∑ denotes polynomials in the lag operator of the exogenous variables,

and = (, … , )′ is a k-dimensional white noise or innovation process; with () = , (′) = ∑ and (′ ) = 0 for " ≠ $. and contain the contemporaneous

(23)

Γ_%() = Γ_&()+ ' (3.2) where Γ_%() = ( − ∑ denotes polynomials in the lag operator of the endogenous variables, is a vector of exogenous variables, Γ_& = + ∑ denotes polynomials in the lag operator of the exogenous variables, and ' is a vector of serially uncorrelated reduced form disturbances. The relationships between the structural equations in (1) and the reduced form in (2) are:

Γ% = ())Γ() = ())* − ∑ + (3.3)

and

' = ()) (3.4)

Hence, the reduced form disturbances are composites of the structural disturbances which give rise to an identification problem. The parameters of the primitive or structural model have to be recovered from the estimated reduced form model, which is done by imposing restrictions on the structural form. As long as the parameters in ()) are not identified, it is impossible to identify the structural disturbances form the reduced form estimates '. The parameters in the structural from equation and those in reduced form equation are related by Γ_%() = ( − ())(∑ )) (3.5) and

Σ-= (''′) = ()) .())/0 0 = ())Σ.())/′ (3.6)

When the VAR includes k variables, full identification of ()) requires 12 restrictions. The covariance matrix Σ_- which describes the pattern of variances in covariances underlying the disturbance terms, provides 1/2k(k+1) of these constraints. This implicates that we need 1/2k(k-1) for exact identification of the structural parameters. When economic theory leads to more imposed restrictions the model is said to be overidentified.

Restrictions on the matrix ()) are known as short-term restrictions because they concern constraints on the disturbances. The most popular restriction is to impose a recursive

identification method by assuming the existence of a Wold-causal chain; disturbances are orthogonalized by the Choleski decomposition. The Bernanke –Sims procedure allows for types of identification restrictions that permit non-recursive structures. They come in various forms: coefficient (sign) restrictions, variance restrictions and symmetry restrictions. See Enders (2010) for a discussion on the application of each of those types of restrictions. An alternative to short-term restrictions are provided by Blanchard and Quah (1989) who separate temporary and permanent effects of disturbances. More specific, they identify demand and supply shocks by assuming that only the latter has a long-run effect on output. The Blanchard and Quah (1989) decomposition is known as a long-run restriction.

Occasionally, a combination of short and long-run restrictions is used in order to achieve identification, see for example Galí (1992).

Crazy Rain and the Little Dragon at Europe’s Door