Dans ce mémoire, nous nous intéressons à une façon géométrique de voir la méthode du Lasso en régression linéaire. Le Lasso est une méthode qui, de façon simultanée, estime les coefficients associés aux prédicteurs et sélectionne les prédicteurs importants pour expliquer la variable réponse. Les coefficients sont calculés à l’aide d’algorithmes computationnels. Malgré ses vertus, la méthode du Lasso est forcée de sélectionner au maximum n variables lorsque nous nous situons en grande dimension (p > n). De plus, dans un groupe de variables corrélées, le Lasso sélectionne une variable “au hasard”, sans se soucier du choix de la variable. Pour adresser ces deux problèmes, nous allons nous tourner vers le Lasso Linéaire. Le vecteur réponse est alors vu comme le point focal de l’espace et tous les autres vecteurs de variables explicatives gravitent autour du vecteur réponse. Les angles formés entre le vecteur réponse et les variables explicatives sont supposés fixes et nous serviront de base pour construire la méthode. L’information contenue dans les variables explicatives est projetée sur le vecteur réponse. La théorie sur les modèles linéaires normaux nous permet d’utiliser les moindres carrés ordinaires (MCO) pour les coefficients du Lasso Linéaire. Le Lasso Linéaire (LL) s’effectue en deux étapes. Dans un premier temps, des variables sont écartées du modèle basé sur leur corrélation avec la variable réponse; le nombre de variables écartées (ou ordonnées) lors de cette étape dépend d’un paramètre d’ajustement γ. Par la suite, un critère d’exclusion basé sur la variance de la distribution de la variable réponse est introduit pour retirer (ou ordonner) les variables restantes. Une validation croisée répétée nous guide dans le choix du modèle final. Des simulations sont présentées pour étudier l’algorithme en fonction de différentes valeurs du paramètre d’ajustement γ. Des comparaisons sont effectuées entre le Lasso Linéaire et des méthodes compétitrices en petites dimensions (Ridge, Lasso, SCAD, etc.). Des améliorations dans l’implémentation de la méthode sont suggérées, par exemple l’utilisation de la règle du 1se nous permettant d’obtenir des modèles plus parcimonieux. Une implémentation de l’algorithme LL est fournie dans la fonction R intitulée linlasso, disponible au https://github.com/yanwatts/linlasso. / In this thesis, we are interested in a geometric way of looking at the Lasso method in the context of linear regression. The Lasso is a method that simultaneously estimates the coefficients associated with the predictors and selects the important predictors to explain the response variable. The coefficients are calculated using computational algorithms. Despite its virtues, the Lasso method is forced to select at most n variables when we are in highdimensional contexts (p > n). Moreover, in a group of correlated variables, the Lasso selects a variable “at random”, without caring about the choice of the variable. To address these two problems, we turn to the Linear Lasso. The response vector is then seen as the focal point of the space and all other explanatory variables vectors orbit around the response vector. The angles formed between the response vector and the explanatory variables are assumed to be fixed, and will be used as a basis for constructing the method. The information contained in the explanatory variables is projected onto the response vector. The theory of normal linear models allows us to use ordinary least squares (OLS) for the coefficients of the Linear Lasso. The Linear Lasso (LL) is performed in two steps. First, variables are dropped from the model based on their correlation with the response variable; the number of variables dropped (or ordered) in this step depends on a tuning parameter γ. Then, an exclusion criterion based on the variance of the distribution of the response variable is introduced to remove (or order) the remaining variables. A repeated cross-validation guides us in the choice of the final model. Simulations are presented to study the algorithm for different values of the tuning parameter γ. Comparisons are made between the Linear Lasso and competing methods in small dimensions (Ridge, Lasso, SCAD, etc.). Improvements in the implementation of the method are suggested, for example the use of the 1se rule allowing us to obtain more parsimonious models. An implementation of the LL algorithm is provided in the function R entitled linlasso available at https://github.com/yanwatts/linlasso.

The ordinary least squares (OLS) is one of the most widely used methods for modelling the functional relationship between variables. However, this estimation procedure counts on some assumptions and the violation of these assumptions may lead to nonrobust estimates. In this study, the simple linear regression model is investigated for conditions in which the distribution of the error terms is Generalised Logistic. Some robust and nonparametric methods such as modified maximum likelihood (MML), least absolute deviations (LAD), Winsorized least squares, least trimmed squares (LTS), Theil and weighted Theil are compared via computer simulation. In order to evaluate the estimator performance, mean, variance, bias, mean square error (MSE) and relative mean square error (RMSE) are computed.

Made available in DSpace on 2016-06-02T20:06:02Z (GMT). No. of bitstreams: 1 2567.pdf: 3332442 bytes, checksum: 1e03b44e1c1f61f90b947fdca5682355 (MD5) Previous issue date: 2009-03-06 / Financiadora de Estudos e Projetos / Several techniques are proposed to determine the lag length of a dynamic regression model. However, none of them is completely satisfactory and a wrong choice could imply serious problems in the estimation of the parameters. This dissertation presents a review of the main criteria for models selection used in the classical methodology and presents a way for determining the lag length from the perspective Bayesian. A Monte Carlo simulation study is conducted to compare the performance of the significance tests, R2 adjusted, final prediction error, Akaike information criterion, Schwarz information criterion, Hannan-Quinn criterion, corrected Akaike information criterion and fractional Bayesian approach. Two estimation methods are also compared, the ordinary least squares and the Almon approach. / Na literatura, muitas técnicas são propostas para determinar o tamanho da defasagem de um modelo de regressão dinâmico. Entretanto, nenhuma delas é completamente satisfatória e escolhas erradas implicam em sérios problemas na estimação dos parâmetros. Este trabalho apresenta uma revisão dos principais critérios de seleção de modelos disponíveis na metodologia clássica, assim como aborda uma maneira de determinar o tamanho da defasagem sob a perspectiva Bayesiana. Um estudo de simulação Monte Carlo é conduzido para comparar a performance dos testes de significância, do R2 ajustado, do erro de predição final, dos critérios de informação de Akaike, Schwarz, Hannan-Quinn e Akaike corrigido e da aproximação Bayesiana fracionada. Também serão comparados os métodos de estimação de Mínimos Quadrados Ordinários e de Almon.

M.Comm. (Financial Management) / There is a wide perception that bank net interest margins (NIMs) in Sub-Saharan Africa in general and South Africa in particular, are higher compared to other regions. The study investigates four commercial banks in South Africa with the aim of identifying the relevant factors affecting the behaviour of NIMs in commercial banking in South Africa, and draws comparisons with other markets. The study employs the Classical Linear Regression Model (CLRM) using the Ordinary Least Squares (OLS) data estimating technique to analyse net interest margins over the period 2000 to 2010. The study takes note of Ho and Saunders’s seminal work produced in 1981, and subsequent extensions and modification by other authors and researchers. Net interest margins are modeled in a single-step together with explanatory variables driven from the theoretical model. Using data obtained from the Bankscope data base, the variables examined in the study are; competitive structure of the market, average operating costs, management’s propensity for risk aversion, credit risk exposure, the quantum of the bank’s operations, short-term money market interest rate volatility, the opportunity cost of holding reserves and quality of management running the institution. The findings of the study suggest that market power, average operating costs, degree of risk aversion, credit risk exposure, and size of operations are major factors explaining the behaviour of NIMs in South Africa. These variables are major in terms of the number of banks that exhibit statistical significance. Market power, interest rate volatility and opportunity cost of holding reserves are also relevant factors, although they affect fewer banks than the major factors. Comparison of South African net interest margins determinants with those from other regions reveals some fundamental differences. These differences indicate that banks from different countries and regions are faced with different operating environments and risk profiles that drive net interest margins.

The main subject of the diploma thesis is a market price modeling by real estates. As a tool for modeling, is used a multiple linear regression. As starting points, are used an econometrical theory and knowledge about real estate valuation. The main goal is to find optimal model for best capture in the time and place.

Access to education is a central part of federal higher education policy, and federal grant and loan programs are in place to make college degrees more attainable for students. However, there is still controversy about whether there are unintended consequences of implementing and maintaining these programs, and whether they are effectively achieving the goal of increased accessibility. In order to answer questions about whether three specific types of federal aid cause higher tuition rates and whether these programs increase graduation rates, four ordinary least squares regression models were estimated. They include changes in both in-state and out-of-state tuition sticker prices, graduation rates, as well as changes in three types of federal aid, and other variables indicative of the value of a degree for four-year public universities in Arizona, California, Georgia, and Florida for years 2001-2011. The regressions indicate a positive effect of Pell Grants on in-state and out-of-state tuition and fees, a positive effect of disbursed subsidized federal loans on the change in number of degrees awarded, and a positive effect of Pell Grants on graduation rates.

Accurately modeling aerodynamic forces and moments are crucial for understanding thebehavior of an aircraft when performing various maneuvers at different flight conditions.However, this task is challenging due to complex nonlinear dependencies on manydifferent parameters. Currently, Computational Fluid Dynamics (CFD), wind tunnel,and flight tests are the most common methods used to gather information about thecoefficients, which are both costly and time–consuming. Consequently, great efforts aremade to find alternative methods such as machine learning. This thesis focus on finding machine learning models that can model the static and thedynamic aerodynamics coefficients for lift, drag, and pitching moment. Seven machinelearning models for static estimation were trained on data from CFD simulations.The main focus was on dynamic aerodynamics since these are more difficult toestimate. Here two machine learning models were implemented, Long Short–TermMemory (LSTM) and Gaussian Process Regression (GPR), as well as the ordinaryleast squares. These models were trained on data generated from simulated flighttrajectories of longitudinal movements. The results of the study showed that it was possible to model the static coefficients withlimited data and still get high accuracy. There was no machine learning model thatperformed best for all three coefficients or with respect to the size of the training data.The Support vector regression was the best for the drag coefficients, while there wasno clear best model for the lift and moment. For the dynamic coefficients, the ordinaryleast squares performed better than expected and even better than LSTM and GPR forsome flight trajectories. The Gaussian process regression produced better results whenestimating a known trajectory, while the LSTM was better when predicting values ofa flight trajectory not used to train the models. / Att noggrant modellera aerodynamiska krafter och moment är avgörande för att förståett flygplans beteende när man utför olika manövrar vid olika flygförhållanden. Dennauppgift är dock utmanande på grund av ett komplext olinjärt beroende av många olikaparametrar. I nuläget är beräkningsströmningsdynamik (CFD), vindtunneltestningoch flygtestning de vanligaste metoderna för att kunna modellera de aerodynamiskakoefficienterna, men de är både kostsamma och tidskrävande. Följaktligen görs storaansträngningar för att hitta alternativa metoder, till exempel maskininlärning. Detta examensarbete fokuserar på att hitta maskininlärningmodeller som kanmodellera de statiska och de dynamiska aerodynamiska koefficienterna för lyftkraft,luftmotstånd och stigningsmoment. Sju olika maskininlärningsmodeller för destatiska koefficienterna tränades på data från CFD–simuleringar. Huvudfokus lågpå den dynamiska koefficienterna, eftersom dessa är svårare att modellera. Härimplementerades två maskininlärningsmodeller, Long Short–Term Memory (LSTM)och Gaussian Process Regression (GPR), samt minstakvadratmetoden. Dessa modellertränades på data skapad från flygbanesimuleringar av longitudinella rörelser. Resultaten av studien visade att det är möjligt att modellera de statiskakoefficienterna med begränsad data och ändå få en hög noggrannhet. Ingen avde testade maskininslärningsmodelerna var tydligt bäst för alla koefficienterna ellermed hänsyn till mängden träningsdata. Support vector regression var bäst förluftmotstånds koefficienterna, men vilken modell som var bäst för lyftkraften ochstigningsmomentet var inte lika tydligt. För de dynamiska koefficienterna presterademinstakvadratmetoden bättre än förväntat och för vissa signaler även bättre än LSTMoch GPR. GPR gav bättre resultat när man uppskattade koefficienterna för enflygbanan man tränat modellen på, medan LSTM var bättre på att förutspå värdenaför en flybana man inte hade tränat modellen på.

Purpose – In recent decades the inflation- and interest rates have followed a long-termdeclining trend. Followed by central banks starting to use unconventional monetary policiesto cope with financial crises have led to increased amounts of liquidity in the financialsystems and available and looking for investment alternatives on the capital markets. At thesame time real estate property prices have set new highs corresponding to a longer-termtrend of declining cap rates. The traditional cap rate formula components the risk-free rateand risk premium less rental growth do not entirely explain the trend of declining cap ratesthat have led to very low cap rates. The purpose with this thesis quantitative study is to testif the newer cap rate determinants money supply and foreign investments percentage of allmarket transactions can explain the decline and recently very low cap rates.Design/Methodology – The master thesis firstly conducts a literature review on previousstudies on cap rate determinants and subsequent conduct an own quantitative study byrunning dynamic ordinary least squares (DOLS) regression analysis on time series ofStockholm central business district commercial office cap rates and determinants chosen asindependent variables representing macro- and market fundamental factors driving caprates with the addition of money supply proxied by monetary aggregate M3 and foreigninvestments on the market proxied by the foreign share of investments on the Stockholmcommercial office market.Findings – The DOLS regression model (1)-(5) determinants are overall significant androbust. Of the newer cap rate determinants that are tested the monetary aggregate M3 andits included lags are of higher and stronger significance and impact on the cap rate than theforeign investment share. However, the foreign investment share time-series data used inthe study do not entirely correspond to the Stockholm central business district (CBD)commercial office market but rather against the larger Stockholm commercial office marketand hence the foreign investment share is assumed to be a relevant cap rate determinantwith support from studies referred to in the section theoretical framework.Research limitations/implications – To ensure high quality in statistical analysis andhypothesis testing large data samples corresponding to longer time-series data than waspossible to obtain for this thesis quantitative study is required. However, even though arelatively small sample has been used it performed well in tests conducted of the dataquality.Originality/value – The master thesis aims to measure and quantify the impact fromunconventional monetary policy and international real estate investments on commercialoffice cap rates. Executed on the Swedish capital Stockholms CBD office market. / Syfte – De senaste decennierna har inflationen- och räntorna följt en långsiktigtnedåtgående trend. Detta har följts upp av att centralbanker börjat använda okonventionellpenningpolitik för att hantera finanskriser, vilket lett till att en ökad penningmängd i definansiella systemen sökt investeringsmöjligheter på kapitalmarknaderna. Samtidigt harfastighetspriserna satt nya rekord ett flertal gånger vilket motsvarar en långsiktig trend avsjunkande kapitaliseringstakter. Den traditionella modellen för kapitaliseringstakt med riskfriränta och riskpremium med subtraktion av hyrestillväxt förklarar inte helt detta. Syftet meddetta examensarbetes kvantitativa studie är att pröva om de nyare potentielladeterminanterna penningmängd och andelen utländska investeringar kan förklaranedgången och de nyligen väldigt låga kapitaliseringstakterna.Design/Metodik – Examensarbetet börjar med en litteraturstudie inom konceptetkapitaliseringstakt och dess determinanter. Därefter görs en kvantitativ studie med DOLSregressionsanalys av kapitaliseringstakter för kontorsmarknaden i Stockholms centralaaffärsdistrikt och modeller av dess determinanter bestående av makro- ochmarknadsfundamentala faktorer som antas vara drivande för kapitaliseringstakten. Medtillägg av determinanterna penningmängd modellerad med måttet M3 och andelenutländska investeringar modellerad med andelen utländska investeringar påkontorsmarknaden i Stockholm.Undersökningsresultat – Kapitaliseringstakts determinanterna i modellerna (1)-(5) ärövergripande signifikanta och robusta. Av de två nyare determinanterna är penningmängdM3 och dess inkluderade laggade värden av högre och starkare signifikans och med störreinverkan på kapitaliseringstakten än andelen utländska investeringar. Dessvärre motsvararinte tidsserien av andelen utländska investeringar på kontorsmarknaden i Stockholm detmindre segmentet kontorsmarknaden för Stockholms centrala affärsdistrikt och därmedantas den motsvarande andelen utländska investeringar vara signifikant med stöd av tidigarestudier som lyfts fram i litteraturstudien.Begränsningar/implikationer – För att säkerställa hög kvalité på analys av statistik behövsstora stickprov vilket motsvarar data för långa tidsserier, vilket inte var möjligt att erhålla fördetta examensarbetes kvantitativa studie. Positivt är dock ändå att tidsserie data somanvänts, trots att de inte är så långa som önskat, presterat bra i de genomförda testerna avdatakvalitén.

Approved for public release, distribution is unlimited / The purpose of this thesis is threefold. Firstly, to attempt to provide validity for the two-sided matching process; secondly, analyze FITREP attributes to determine their suitability for a weighted criteria evaluation system and; thirdly, compare the USMC promotion and assignment process with contemporary human resource management practices. Using data from the USMC Officer Accession Career file (MCCOAC), a logit model is used to estimate the effects of TBS preference and other officer characteristics on retention to the seven year mark. Findings indicate that there was little difference in the probability of retention throughout most preference levels except for the bottom sixth. Using USMC FITREP data, an ordinary least squares model is used to estimate the effects of rank and MOS on FITREP scores across all attributes. Multiple comparison tests demonstrated that there are statistical differences at the 0.05 level between the means of the MOSs. Additionally, reporting creep is continuing across all attributes. Surveys were also conducted. The first survey indicated that USMC officers believe the FITREP attributes were not all equally important within, and across each MOS - although the USMC assesses them as such. The second survey indicated that the USMC promotion and assignment process can be strengthened through a clearly defined HRM plan that extends beyond 'faces' and 'places', and provides very clear links to the organizational strategy. Based on the findings it is recommended that the USMC review its HRM processes and conduct further analyses on the FITREP data for: (1) correlation, (2) longitudinal analysis as a predictor for success and, (3) relevance and relationship to MOS characteristics, position descriptions, and organizational strategy. / Major, Royal Australian Infantry Corps / Major, United States Marine Corps

This study aimed to assess the determinants and deterrents of profit shifting, which can occur as a result of corporate income tax competition, with a view to aid in collecting sufficient tax revenue to meet public spending requirements. The study theoretically and empirically analysed the effectiveness of the introduction of the South African transfer pricing regulations on deterring the occurrence of profit shifting in South Africa using annual financial information of South African parented multinational enterprises for the period 2010 – 2017. The study established that the implementation of transfer pricing regulations resulted in a reduction in profit shifting that became increasingly more prominent as the rules became stricter. Based on the findings of the study, it is recommended that the South Africa government should allocate sufficient resources to ensure that the transfer pricing regulations are being adhered with an aim to reduce profit shifting from South Africa. / Economics / M. Com. (Economics)