Este estudo versa sobre uma investigação de viabilidade da utilização de redes neurais auto-organizadas na classificação e exploração de dados macroeconômicos. Para tanto, foi elaborado um método no qual foram empregadas topologias neurais auto-organizadas, procurando assim explorar as características de melhor desempenho de cada um dos modelos, sob um enfoque seqüencial e com o intuito de se adquirir conhecimento intermediário em cada uma de suas fases, diminuindo o impacto da complexidade tanto no tempo requerido para realização da tarefa quanto na análise dos resultados. Os próprios resultados obtidos sugerem que a utilização de redes neurais artificiais auto-organizadas na aquisição de conhecimento sobre bases de dados aplicáveis às Ciências Econômicas apresenta desempenho análogo aos modelos paramétricos tradicionalmente empregados na construção de cenários com tais informações. / This study turns on an inquiry of viability of the use of self-organizing neural nets in classification and exploration of macroeconomic data. For this purpose, a method in which had been used self-organized neural topologies was elaborated, looking to explore the better characteristics of performance of each one of the models, under a sequential approach and with objective of acquiring intermediate knowledge in each one of its phases, diminishing the impact of the complexity as in time consuming as in analysis of results. Main results obtained suggest the use of self-organized artificial neural nets in acquisition of knowledge on Economic databases presents analog performance to traditional parametric models in scenarios building.

Este estudo versa sobre uma investigação de viabilidade da utilização de redes neurais auto-organizadas na classificação e exploração de dados macroeconômicos. Para tanto, foi elaborado um método no qual foram empregadas topologias neurais auto-organizadas, procurando assim explorar as características de melhor desempenho de cada um dos modelos, sob um enfoque seqüencial e com o intuito de se adquirir conhecimento intermediário em cada uma de suas fases, diminuindo o impacto da complexidade tanto no tempo requerido para realização da tarefa quanto na análise dos resultados. Os próprios resultados obtidos sugerem que a utilização de redes neurais artificiais auto-organizadas na aquisição de conhecimento sobre bases de dados aplicáveis às Ciências Econômicas apresenta desempenho análogo aos modelos paramétricos tradicionalmente empregados na construção de cenários com tais informações. / This study turns on an inquiry of viability of the use of self-organizing neural nets in classification and exploration of macroeconomic data. For this purpose, a method in which had been used self-organized neural topologies was elaborated, looking to explore the better characteristics of performance of each one of the models, under a sequential approach and with objective of acquiring intermediate knowledge in each one of its phases, diminishing the impact of the complexity as in time consuming as in analysis of results. Main results obtained suggest the use of self-organized artificial neural nets in acquisition of knowledge on Economic databases presents analog performance to traditional parametric models in scenarios building.

The thesis focuses on finding solutions to major exchange rate puzzles, which were discussed in detail by Obstfeld and Rogoff (2000). The first puzzle is the purchasing power parity puzzle. The first version of the latter puzzle is concerned with whether a real exchange rate reverts in the mean. To resolve the puzzle in the context of Southern African Development Community countries, the thesis uses Bayesian unit root testing and nonlinear nonstationarity tests associated with the smooth transition autoregressive family of models. According to Bayesian unit root test results, the nonstationarity hypothesis received small posterior probability relative to other hypotheses. In this setting, the Bayesian results strongly supported the hypothesis that all the real exchange rates were trend-stationary autoregressive processes. However, it should be pointed out that Ahking (2004) has found these tests to be biased toward trend stationarity. Nonlinear nonstationarity tests presented evidence that four out of ten of SADC’s real exchange rates could be regarded as nonlinear globally ergodic processes, while others could be considered random walks. The thesis relies on local-to-unity asymptotic theory and Rossi (2005a) to deal with the half-life version of the PPP puzzle. The half-life version is that a high degree of exchange rate volatility is generally associated with an implausibly slow speed of mean reversion. Depending on the robustness of the methods used, empirical evidence points to several half-lives of less than 36 months, but the confidence intervals of half-life deviations from PPP are found in all cases, as in Rossi’s work, to be too wide to be informative enough to resolve the puzzle. In addition, the thesis undertakes Hinich and Chong (2007) class tests of fractional integration to ensure that a long memory process is not mistaken for a nonstationary process in finding solutions to the PPP puzzle. The results show that at 1 per cent and 5 per cent significance levels, the real exchange rates associated with South Africa, Mauritius and Swaziland are not fractionally integrated. Tanzania’s real exchange rate was found to be stationary-fractionally integrated but with the antipersistence property. Other currencies were found to be nonstationary-fractionally integrated. The third puzzle is the exchange rate determination puzzle, which is as follows: In the short run there seems to be no reliable determinants of exchange rates. The thesis relies on the market microstructure approach to find the determinants of South Africa’s exchange rate. In this context, the thesis utilises autoregressive distributed lag model of cointegration to identify the fundamental and non-fundamental determinants of the rand/dollar exchange rate. The main contribution of the thesis to the economic literature is the usage of newly developed methods in an attempt to resolve the above-mentioned puzzles. / Thesis (PhD (Economics))--University of Pretoria, 2008. / Economics / unrestricted

本論文探討利用總體因子預測台灣股市的月成交量，並討論其預測準確度。總體因子主要利用主成分分析法從大量的總體資料中抽出，台灣股市月成交量資料主要來自TEJ資料庫，並將其分為九類：水泥窯業、食品業、塑膠化工業、紡織業、機電業、造紙業、營建業、金融業和加權指數。 結果發現三個月後的預測值比一個月後的預測值準確，而從RMSE跟MAE的結果，發現食品業、塑膠化工業、紡織業、機電業、造紙業預測的準確度較高。 / This paper discusses forecasting monthly turnover by static principle components method, and testing accuracy of forecasting. The monthly turnover is from Taiwan stock market as nine turnover classification, Cement & Kiln industry, Food industry, Plastic & Chemical industry, Textile industry, Mechanical & Electrical industry, Paper-making industry, Construction industry, Financial industry and Value-Weighted Index. The principle components extracted from large macroeconomic datasets have the explanatory power to monthly turnover. In addition, for basic forecasting, the accuracy of three-month prediction is better than one-month prediction in both subsamples. To test accuracy, RMSE (PC) and MAE (PC) are outperformed the same in Food industry, Textile& Fibers industry. However, MAE (PC) in Plastic & Chemical industry, RMSE (PC) in Mechanical & Electrical industry and Paper-making industry still show the good prediction as well.

A wide variety of machine learning algorithms havebeen used to predict stock prices. The aim of this project hasbeen to implement a machine learning algorithm using supportvector regression to predict the stock price of two well knowncompanies—Apple and Microsoft—one day into the future usingthe current day’s stock price, macroeconomic data and microeconomicdata and to compare the prediction error with the differentdata inputs. The results show that the addition of macroeconomicand microeconomic data did not improve the prediction error.This suggests that the macroeconomic and microeconomic dataused in this project does not contain additional information aboutfuture stock prices. The results also show that support vectorregression performs worse than linear regression, however inthis case no definite conclusion can be drawn since only onekernel and a handful of parameter values were considered whentraining and testing the algorithm. However, these results mightalso suggest that using the current day’s data is not sufficient tobe able to predict the non-linear relationships. / Ett flertal maskininlärnings-algoritmer har använts för att förutspå aktiepriser. Målet med det här projektet har varit att implementera en maskininlärnings-algoritm som använder sig av support vector regression för att förutspå aktiepriset av två välkända företag—Apple och Microsoft—en dag in i framtiden genom att använda dagens aktiepris, makroekonomisk data och mikroekonomisk data samt att jämföra prediktionsfelet med dem olika indata. Resultaten indikerar att additionen av makroekonomisk och mikroekonomisk data inte förbättrade prediktionsfelet. Detta antyder att den makroekonomiska och mikroekonomiska data som användes i projektet inte innehåller någon ytterliggare information om framtida aktiepriser. Resultaten indikerade också att linjär regression presterar bättre än support vector regression, men i detta fallet kan ingen definitiv slutsats dras eftersom endast en kernel och ett par parameter-värden användes för att träna och testa algoritmen. Däremot kan dessa resultat också antyda att a inte är tillräcklig för att kunna förutspå dem icke-linjära förhållandena. / Kandidatexjobb i elektroteknik 2021, KTH, Stockholm

Popularity of nonlinear threshold models and unit root tests has increased after the recent empirical studies concerning the effects of business cycles on macroeconomic data. These studies have shown that an economic variable may react differently in response to downturns and recoveries in a business cycle. Inspiring from empirical results, this thesis investigates dynamics of Turkish key macroeconomic data, namely capacity utilization rate, growth of import and export volume indices, growth of gross domestic product, interest rate for cash loans in Turkish Liras and growth of industrial production index. Estimation results imply that capacity utilization rate and growth of industrial production index show M-TAR type nonlinear stationary behavior according to the unit root test proposed by Enders and Granger (1998).