Spring on the York College campus

Business Administration

Add discipline info here.

  • Business Analytics: Andrew Rhodes

    Data Mining Techniques to Identify Credit Risk

    Researcher: Andrew Rhodes

    Faculty Mentor: Dr. Wei Chen

    Abstract: Using 1000 records of credit data, detailing 29 variables of interest to a bank of a prospective borrower, we look to understand the ability of data mining techniques to predict these credit decisions. The total sample space was partitioned into randomly selected 60% train, 25% validation and 15% test sets. Each of the three models of interest, a logistic regression model, K-nearest neighbor model and neural network model were then specified using the training partition. Once fit, the models were each tested on the validation partition to select the model of greatest accuracy. The logistic regression model is found to have the greatest amount of accuracy and therefore is the model of choice. When the model was tested via the test partition, it was found to have an accuracy of 0.80. In a scenario in which the bank experiences a 5-to-1 loss on bad loans approved vs. good loans approved, none of the three models were profitable. Therefore, we then analyze the probit function of the logistic regression model to understand what propensity of success that should act as the threshold over which the bank should extend credit to maximize profit. This analysis suggests that the bank should approve all loans that return a logistic probit propensity of 0.799 or greater, as this returned the highest net profit on the validation partition assuming a 5-to-1 loss-to-reward ratio. This propensity of success existed in applicants of the 43rd percentile of the validation set.

  • Business Analytics: Kylee Fisher

    Applying Data Mining Techniques in German Credit Applications

    Researcher: Kylee Fisher

    Faculty Mentor: Dr. Wei Chen

    Abstract: Money-lending is the world’s second oldest profession but it wasn’t until the beginning of the 20th century that credit companies were founded to share information about credit. Credit agencies often collect large amounts of data to predict when defaults or similar events can occur. This research investigates applying known data mining techniques to a German Credit dataset to investigate if those techniques can achieve equivalent, or better, results in determining whether applicants have “good credit” or “bad credit” as the results attained before predictive modeling.

  • Economics: Andrew Rhodes

    An Analysis of Dynamic Regression Modeling of Nonlinear Time Series: Vector Auto

    Researcher: Andrew Rhodes

    Faculty Mentor: Dr. Nicole Sadowski

    Abstract: Attempts to forecast gross domestic product for many economies across the globe are prominent among available literature. In recent history, econometricians have utilized the availability of computers and algorithmic functionality to build models that capture the effect that the selected variables have on gross domestic product. Some models look to employ a simple linear regression to characterize the time series and compute expected values for the economic data of interest, however these models typically fall short in capturing the non-linear movement of econometric time series. The advancements made by Hamilton (1989) through his specification of a Markovian regime switching model allows researchers to explore dynamic regression models that identify the various states economic data may exist and the path of the time series through these states. These models, although robust, fail to consider the interactivity of several variables among each other that often exists in a complex economy and, unless captured in forecasting systems, are not capable of correctly specifying the behavior of the time series of interest. This paper looks to explore indicators used in the calculation of the Conference Board’s Leading Economic Index® and real gross domestic product of the United States throughout various econometric models—beginning with a simple linear regression model, to a Markovian regime-switching model and finally ending with a vector autoregression model. The VAR model provides interesting insight into the interdependence of economic time series and allows the researcher to explore the casual link between these time series. Finally, an impulse response and forecast error variance decomposition model are created and examined to understand the sensitivity of each of these time series on one another.