Search

Elektrik; sürdürülebilir yaşamda önemli bir rol oynayan ve çeşitli sektörlere katma değeri çok yüksek olan enerji türüdür. Elektrik, sosyo-ekonomik kalkınmada stratejik önemde bulunduğu için ekonomik refahın ve büyümenin en önemli aktörlerindendir. Yapısı gereği depolanamayan ve üretildiği anda tüketilmesi gereken bu enerji türü, ekonomik kalkınmanın tüm yönleri ile entegre olması ve aynı zamanda tek bir modelin her zaman doğru tahminleri vermemesi sebebi ile elektrik talep tahmini çalışmaları her dönem güncelliğini korumuştur. Dolaysıyla bu tez çalışmasında elektrik talep tahmini farklı yöntem ve modeller ile tahmini gerçekleştirilmiştir. Tez çalışmasında, uygulama dönemi, 2007:01-2020:12 belirlenmiştir. Bu dönemin belirlenmesindeki en önemli sebep ise ekonomi ve sosyal hayatta yaşanan birtakım olumsuzlukların bulunmasıdır. Çalışma, tek değişkenli ve çok değişkenli olmak üzere iki farklı uygulama ile gerçekleştirilmiştir. Tek değişkenli modellerde, brüt elektrik talep miktarı kullanılırken, çok değişkenli model çalışmalarında ise brüt elektrik talep miktarı, tüketici fiyat endeksi, sanayi üretim endeksi, ülkeye gelen turist sayısı ve işsizlik değişkenleri kullanılmıştır. Zaman serisi modellerine ilişkin uygulamalar Eviews 10 paket programı ile gerçekleştirilirken yapay sinir ağı ve hibrit yöntem uygulamaları MATLAB ile yapılmıştır. Yapılan uygulama sonuçlarında, resmi makamlarca açıklanan 2021:01-2021:10 dönemi talep miktarı ile tek değişkenli ve çok değişkenli modeller ile tahmin edilen talep miktarları karşılaştırılmış ve istatistiksel performans kriterlerine göre en düşük hata değerlerine sahip olan model çok değişkenli yapay sinir ağı mimarisi olmuştur. Çalışma bu noktadan sonra 2022:07 dönemine kadar talep tahmini gerçekleştirilmiş ve çok değişkenli yapay sinir ağı mimarisi ile hibrit yöntem benzer dalgalanmalar sergilemiştir. Bu dönem için çok değişkenli yapay sinir ağına göre 28519.12993 GWh olarak tahmin edilirken tek değişkenli yapay sinir ağına göre ise 27009.25479 GWh tahmin edilmiştir. Electricity is a type of energy that plays an important role in sustainable life and has a very high added value in changing sectors. Electricity is one of the most important actors in economic prosperity and growth, as it has strategic importance in socio-economic development. This type of energy, which cannot be stored due to its nature and must be consumed as soon as it is produced, has always been up-to-date in electricity demand forecasting studies since it is integrated with all aspects of economic development, and at the same time, a single model does not always give accurate forecasts. Therefore, in this thesis, electricity demand forecasting was carried out with different methods and models. The application period for the thesis study is January 2007–December 2020. The most important reason for determining this period is the existence of some negativities in economic and social life. The study was carried out with two different applications as univariate and multivariate. In univariate models, gross electricity demand amount is used, while in multivariate model studies, gross electricity demand amount, consumer price index, industrial production index, number of tourists coming to the country, and unemployment variables are used. While applications related to time series models were carried out with Eviews 10 package program, artificial neural network and hybrid method applications were made with MATLAB. In the results of the application, the demand amount for the period January 2021– November 2021 announced by the official authorities and the estimated demand amounts with univariate and multivariate models were compared, and the model with the lowest error values according to statistical performance criteria was the multivariate artificial neural network architecture. After this point, demand forecasting was carried out until July 2022, and the multivariate artificial neural network and hybrid method exhibited similar fluctuations. For this period, it was forecast as 28519.12993 GWh according to the multivariate artificial neural network, while it was forecast as 27009.25479 GWh according to the univariate artificial neural network.

Inflation refers to an ongoing and overall comprehensive increase in the overall level of goods and services price in the economy. Today, inflation, which is attempted to be kept under control by central banks or, in the same way, whose price stability is attempted, consists of continuous price changes that occur in all the goods and services used by the consumers. Undoubtedly, in terms of economy, in addition to the realized inflation, inflation expectations are also gaining importance. This situation requires forecasting the future rates of inflation. Therefore, reliable forecasting of the future rates of inflation in a country will determine the policies to be applied by the decision-makers in the economy. The aim of this study is to predict inflation in the next period based on the consumer price index (CPI) data with two alternative techniques and to examine the predictive performance of these two techniques comparatively. Thus, the first of the two main objectives of the study are to forecast the future rates of inflation with two alternative techniques, while the second is to compare the two techniques with respect to statistical and econometric criteria and determine which technique performs better in comparison. In this context, the 9-month inflation in April-December 2019 was forecast by Box-Jenkins (ARIMA) models and Artificial Neural Networks (ANN), using the CPI data which consist of 207 data from January 2002 to March 2019 and the predictive performance of both techniques was examined comparatively. It was observed that the results obtained from both techniques were close to each other.

The prediction of an economic crisis is the most critical area of study for all actors related to the economy. Crises, a sign of uncertainty, do not have a specific timeline, but they can be predicted by analyzing particular indications. Studies on predicting the crisis are commonly related to macroeconomic variables. This study addresses an alternative approach to predicting crisis periods, which involves analyzing changes in the trading volumes of companies listed on Borsa Istanbul (BIST) instead of relying solely on macroeconomic variables. The study aims to examine the transaction volume data from 169 firms that regularly traded in BIST between 2000 and 2018. The predictability of economic crises in Türkiye has been investigated by applying binary logistic regression analysis, a methodology commonly employed in the literature as a signal approach for detecting economic crises. Some statistically significant parameters were discovered positive, and some were found negative in estimated logistic regression models, and the companies to which the statistically insignificant parameters belonged were evaluated as companies that did not give a signal for the economic crisis model. The findings suggest that changes in the trading volume of many companies, not just a few ones, can be a valuable predictor of crises.

Bu çalışmada, 01.01.2020-29.12.2020 dönemine ilişkin CDS, BIST100 ve VIX değişkenlerinin günlük getiri serilerinden yararlanılmıştır. ARCH etkisi gösteren en uygun ARMA ve çoklu regresyon modelleri seçilerek, CDS değişkeninin oynaklığını etkileyen faktörleri belirlemek amacıyla, çeşitli simetrik ve asimetrik koşullu değişen varyans (ARCH) modelleri tahmin edilmiştir. Hem ARMA hem de çoklu regresyon ile oluşturulan modellerin varyans denkleminde BIST100’ün bulunduğu modellerin tamamında, bu değişkenin CDS getiri serisinin oynaklığında negatif yönde etki gösterdiği, BIST100 değişkeninin parametresinin, GARCH modellerinde anlamlı bulunduğu, CDS risk primine şok etkilerin uzunluğunun benzer değerlerde [yaklaşık yarım gün: ARMAGARCH (0.5446) ve Çoklu regresyon-GARCH (0.5208)] olduğu ve CDS getiri serisinin oynaklığının çoklu regresyon yerine ARMA modeliyle daha iyi sonuç verdiği belirlenmiştir. In this study, daily return series of CDS, BIST100 and VIX variables for the period 01.01.2020 - 29.12.2020 were used. By choosing the most appropriate ARMA and multiple regression models showing ARCH effect, various symmetric and asymmetric conditional variable variance (ARCH) models were estimated in order to determine the factors affecting the volatility of the CDS variable. In all models with BIST100 in the variance equation of both ARMA and multiple regression models, this variable has a negative effect on the volatility of the CDS return series, the parameter of the BIST100 variable is significant in the GARCH models, the length of the shock effects on the CDS risk premium has similar values [about half a day: ARMA-GARCH (0.5446) and Multiple regressionGARCH (0.5208)] and the volatility of the CDS return series gave better results with the ARMA model instead of multiple regression.