Italy's Economic Crisis: Action Needed to Strengthen Growth Prospects - Research Paper

Introduction

The Italian economy underwent a period of economic crisis, and recently the economy found itself in the center of deep economic disaster. To manage the recent crisis, Italy adopted intensive reform agenda, which promoted more competition, followed by economic growth (Bento & Moutinho, 2016). However, further economic strategies are needed, including further action to strengthen its growth prospects. In this economic scenario, the environmental policy debate has been put aside, and Italy has not yet clearly articulated its future energy strategies. While energy is a key input in the production process, the energy sector remains a primary contributor to environmental pollution. Countries are not ready to compromise economic growth, even amid serious issues of environmental pollution, hence the call to have a proper balance between environmental conservation and economic growth.

Is your time best spent reading someone else’s essay? Get a 100% original essay FROM A CERTIFIED WRITER!

Save 25%

World Health Organization conducted the first conference on Air Pollution and Health in 2018 (Bento & Moutinho, 2016). In the meeting, one of the major observations is that strategic management of the energy sector is highly relevant to have good air quality. In this context, there is an attempt to move to the use of clean energy. Given this background, the research seeks to examine the relationship between GDP growth and Carbon Dioxide emissions in Italy, considering the data from 1960 up to 2016. Also, the research aims at developing two models; Batch Gradient Descent and Stochastic Gradient Descent machine learning algorithms to optimize the prediction of Carbon Dioxide emissions by taking Gross Domestic product as an input variable and Co2 emissions as the output variable.

Literature Review

Researchers have contributed to the literature, based on the study of the relationship between carbon dioxide emissions and economic activity by using a different but complementary statistical approach, having as a focus the investigation of the economic trend and conditions of the Italian environmental conditions (Zhang & Wang, 2013). More imperatively, the recent economic situation provided an opportunity for Italy to reorganize its economy, including considering other substitute sources to fill the gap in its energy requirements. To this end, Italy has restricted local energy resources with a high dependence on outside energy supply (Bouznit & Pablo-Romero, 2016). So an energy import dependency of 52.7% from 1990 against 83.8% in 2010, final energy consumption has been growing steadily, with households, transport and the industry is the greatest energy-consuming sectors (Chen & Huang, 2013). Italian per capita CO2 emissions are way below the 27 average. However, the energy intensity in Italy is lower than the 27 average, but the carbon dioxide emissions are more than the 27 mean levels (Chen & Huang, 2013). In particular, the Italian oil and gas shares in primary energy supply are above the European average, while hydroelectricity and other renewable sources still play a minor role. Given this scenario, it appears the need for Italy to start investing in a significant reduction of CO2 emissions as a priority, before implementing new environmental policy interventions.

Over the last years, Italy has committed to reducing overall greenhouse gas emissions by 13% compared to 1990 levels, increase the share of renewable energy sources in final energy consumption to 17% and cut energy consumption by 27.90 mtoe (Haseeb & Azam, 2015). The study examined the relationship between economic growth and carbon dioxide emissions in Italy, considering the yearly data through 1960 -2016 (Haseeb & Azam, 2015). To this end, the study adopts the Gradient Descent machine learning technique to establish the relationship between Co2 emissions and GDP growth.

For instance, a study was conducted to investigate the relationship among Co2emissions, renewable and non-renewable energy use in the context of Pakistan, in which the study employed the Autoregressive lag model to assess the long-run relationship among the variables (Begum, Sohag, Abdullah & Jaafar, 2015). The result found out that renewable energy consumption is not contributing to CO2 emissions, and non-renewable energy is contributing to the CO2 emissions. The results indicated that in an urban area, solid fuels have a significant contribution to CO2 emissions. The study also found that the existence of environmental alternative energy sources leads to improved environmental quality (Lacheheb, Rahim & Sirag, 2015). A study has also applied the co-integration for 107 low and high-income countries to assess the relations among energy consumption, carbon emissions, and GDP growth (Magazzino, 2016). The result revealed that in low-income countries, renewable energy consumptions increase CO2 emissions, but reduces production. In another empirical study validated the relationship between renewable energy consumption and GDP growth, and utilized Auto-regressive distributive lag and Vector error correction model.

Another study used Artificial Neural Networks to estimate the GDP for given input variables. The study proved that using ANN could improve the accuracy of prediction (Magazzino, 2016). When multiple linear regression, Support Vector Regression, and Artificial Neural Networks are employed in estimating the natural gas usage, the result indicated that using Support vector regression produces the least error when predicting the natural gas usage. Similarly, the use of linear regression in Machine learning can assess the linkages between natural gas usage and seasonal variables (Joo, Kim & Yoo, 2015). The study shows that recurrent neural networks and linear regression models resulted in inaccurate predictions.

More importantly, another study used the Panel Vector Auto-regression and system generalized method of moments in conducting an empirical analysis of 116 countries (Magazzino, 2016). The study found that economic growth does not lead to energy consumption. According to the study, in all regions apart from Caribbean-Latin America, GDP growth does not have a direct impact on CO2 emissions (Riti, Song, Shu & Kamah, 2017).

Data and Methodological framework

Data and Its Description

The study considers the yearly data of Italy from 1960 to 2016; the analysis uses Co2 emissions and GDP growth as dependent and independent variables, respectively. The GDP data is expressed in millions of euros. The research developed one equation for the prediction of Carbon dioxide emissions. The equation predicts the output variable, Carbon dioxide emissions, by taking Gross Domestic product as the input variable.

The Gradient Descent Algorithm

Gradient descent is an iterative optimization algorithm to find the minimum of a function. Here that function is our loss function of a linear regression. The Gradient descent optimization algorithm is the main component in training the predictive model. There are two metrics for Gradient descent optimizer for its performance (1) generalization (model's prediction for out of sample data) and (2) time took for converging within the tolerance limit. This study used two models; Stochastic Gradient Descent and Batch Gradient Descent.

Batch Gradient Descent

For implementing the Batch Gradient Descent, all the training data is taken into consideration to take a single step. The research took the average of the gradients of all the training samples and then used that mean gradient to update the parameters - that is, one step of gradient descent in one epoch. Consequently, the study expected the graph of cost against epochs to be quite smooth because the gradients of training data for a single step were being averaged.

Stochastic Gradient Descent

In stochastic Gradient Descent, the study has considered just one sample at a time to take a single step. The cost fluctuates over the training samples and does not necessarily decrease. However, in the end, the cost starts to decrease with fluctuations. However, SGD converges faster when the dataset is larger.

Predictive Model Building Procedure

Step 1

In this step, the study collected the data for the model, as the process also involves the ingestion of the data into python software for processing and analysis.

Step 2

The research separates the data into two. The input variable (Gross Domestic product growth), and the output variable (Carbon Dioxide emissions)

Step 3

In the third step, an exploratory analysis of the input and output variables have been conducted to plot scatter to assess the relationship between the two variables. Further, both independent and dependent variables are split at a ratio of 80: 20 for training and testing of the model.

Step 4

In this step, training and testing losses were plotted against epochs at around epoch number 60 training. At this iteration number, validation losses are not decreasing further as the epochs are increased so that the ideal epoch would be 60 in this case.

Step 5

The best performing model between SGD and BGD was selected through observing the prediction results based on the root mean squared error (RMSE).

Results and Discussion

Relationship Between Input Variable (GDP) And the Output Variable (CO2)

The research describes a scatter plot of the Output variable (CO2) against the input variable (GDP). The graph below shows a non-linear relationship between the two variables suggesting the need to use more advanced machine learning algorithms that can learn non- linear and more sophisticated mathematical mapping.

Figure 1: Illustrating the scatter plot GDP against Co2 emission from 1960 to 2016

CO2 Model Prediction

The prediction loss graph indicates that the training loss is not decreasing beyond 60th iteration for both the Batch Gradient and Stochastic Gradient optimizers. Therefore, the study took 60 as the ideal epoch for further prediction of the model.

Figure 2: Illustrates Stochastic and Batch Gradient Optimization

The parameters of Batch Gradient Descent and Stochastic Gradient Descent algorithms for predicting the output variable (CO2) are presented in table 2. In both models, 80% of the data are used for model training, and 20% used for model testing. The study compares the prediction strengths of the models in terms of root mean squared error; the lower the RMSE, the better the prediction capacity of the model. From the results, the Batch Gradient Descent (BGD) is having a higher predictive capacity than the Stochastic Gradient Descent algorithm.

Figure 3: Illustrates RMSE performance for Batch and Stochastic Optimization

Conclusion

In this study, a linear regression with a Gradient Descent machine learning algorithm has been used. The research employed both Stochastic and Batch Gradient Descent algorithms in the prediction of Carbon dioxide emissions in Italy with GDP growth as the input variable using annual data through 1960- 2016. More importantly, the research has optimized the prediction of Carbon dioxide emissions by using GDP growth as the input variable. According to the study findings, the Batcher Gradient Descent optimizer performed better than Stochastic Gradient Descent for prediction purposes. Additionally, the exploratory analysis result indicates non- linear relationship within the two variables. Therefore, the study suggests the use of more advanced mac...