TMTatjana Miljkovic

Tatjana Miljkovic

*This is an algorithmically generated author record based on publicly available information.

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article


Tatjana Miljkovic has a total of 26 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Open Access

    Modeling the Inter-Arrival Time Between Severe Storms in the United States Using Finite Mixtures

    by Ilana Vinnik, Tatjana Miljkovic

    When inter-arrival times between events follow an exponential distribution, this implies a Poisson frequency of events, as both models assume events occur independently and at a constant average rate. However, these assumptions are often violated in real-insurance applications. When the rate at which events occur changes over time, the exponential distribution becomes unsuitable. In this paper, we study the distribution of inter-arrival times of severe storms, which exhibit substantial variability, violating the assumption of a constant average rate. A new approach is proposed for modeling severe storm recurrence patterns using a finite mixture of log-normal distributions. This approach effectively captures both frequent, closely spaced storm events and extended quiet periods, addressing the inherent variability in inter-event durations. Parameter estimation is performed using the Expectation–Maximization algorithm, with model selection validated via the Bayesian information criterion (BIC). To complement the parametric approach, Kaplan–Meier survival analysis was employed to provide non-parametric insights into storm-free intervals. Additionally, a simulation-based framework estimates storm recurrence probabilities and assesses financial risks through probable maximum loss (PML) calculations. The proposed methodology is applied to the Billion-Dollar Weather and Climate Disasters dataset, compiled by the U.S. National Oceanic and Atmospheric Administration (NOAA). The results demonstrate the model’s effectiveness in predicting severe storm recurrence intervals, offering valuable tools for managing risk in the property and casualty insurance industry.


  • Open Access

    A dimension reduction assisted credit scoring method for big data with categorical features

    by Tatjana Miljkovic, Pei Wang

    In the past decade, financial institutions have invested significant efforts in the development of accurate analytical credit scoring models. The evidence suggests that even small improvements in the accuracy of existing credit-scoring models may optimize profits while effectively managing risk exposure. Despite continuing efforts, the majority of existing credit scoring models still include some judgment-based assumptions that are sometimes supported by the significant findings of previous studies but are not validated using the institution’s internal data. We argue that current studies related to the development of credit scoring models have largely ignored recent developments in statistical methods for sufficient dimension reduction. To contribute to the field of financial innovation, this study proposes a Dimension Reduction Assisted Credit Scoring (DRA-CS) method via distance covariance-based sufficient dimension reduction (DCOV-SDR) in Majorization-Minimization (MM) algorithm. First, in the presence of a large number of variables, the DRA-CS method results in greater dimension reduction and better prediction accuracy than the other methods used for dimension reduction. Second, when the DRA-CS method is employed with logistic regression, it outperforms existing methods based on different variable selection techniques. This study argues that the DRA-CS method should be used by financial institutions as a financial innovation tool to analyze high-dimensional customer datasets and improve the accuracy of existing credit scoring methods.


  • Clustering regions with dynamic time warping to model obesity prevalence disparities in the United States

    by Katherine Vorpe, Sierra Hessinger, Rebekah Poth, Tatjana Miljkovic

    Current methods for clustering adult obesity prevalence by state focus on creating a single map of obesity prevalence for a given year in the United States. Comparing these maps for different years may limit our understanding of the progression of state and regional obesity prevalence over time for the purpose of developing targeted regional health policies. In this application note, we adopt the non-parametric Dynamic Time Warping method for clustering longitudinal time series of obesity prevalence by state. This method captures the lead and lag relationship between the time series as part of the temporal alignment, allowing us to produce a single map that captures the regional and temporal clusters of obesity prevalence from 1990 to 2019 in the United States. We identify six regions of obesity prevalence in the United States and forecast future estimates of obesity prevalence based on ARIMA models.


  • Open Access

    The Automated Bias-Corrected and Accelerated Bootstrap Confidence Intervals for Risk Measures

    by Bettina Grün, Tatjana Miljkovic

    Different approaches to determining two-sided interval estimators for risk measures such as Value-at-Risk (VaR) and conditional tail expectation (CTE) when modeling loss data exist in the actuarial literature. Two contrasting methods can be distinguished: a nonparametric one not relying on distributional assumptions or a fully parametric one relying on standard asymptotic theory to apply. We complement these approaches and take advantage of currently available computer power to propose the bias-corrected and accelerated (BCA) confidence intervals for VaR and CTE. The BCA confidence intervals allow the use of a parametric model but do not require standard asymptotic theory to apply. We outline the details to determine interval estimators for these three different approaches using general computational tools as well as with analytical formulas when assuming the truncated Lognormal distribution as a parametric model for insurance loss data. An extensive simulation study is performed to assess the performance of the proposed BCA method in comparison to the two alternative methods. A real dataset of left-truncated insurance losses is employed to illustrate the implementation of the BCA-VaR and BCA-CTE interval estimators in practice when using the truncated Lognormal distribution for modeling the loss data.


  • Open Access

    Modeling the Economic Cost of Obesity Risk and Its Relation to the Health Insurance Premium in the United States: A State Level Analysis

    by Thomas Woods, Tatjana Miljkovic

    We propose a new approach for estimating the state-level direct and indirect economic cost of obesity in the United States for the time period 1996 to 2018. Our unique top-down methodology integrates a prevalence-based method with various medical-level costs, economic, demographic, and socio-economic factors. Using this approach, we investigate the relationship between the estimates of the total obesity-related costs and the health insurance premium by state in order to evaluate the state burden of obesity. Our estimate of the total national economic cost attributed to obesity is approximately $422 billion in 2018, representing about 2% of the national GDP for the same year. Using exponential smoothing models, we forecast that the total cost would reach $475 billion in 2021 without accounting for the impact of COVID-19 on obesity. The top states driving the cost estimates are California, Texas, New York, and Florida. A bootstrapping technique is employed to the state-level estimated cost in order to determine the average cost per person. We hope that our study will promote interest in this topic and open discussion for further research in this area.


  • Using Model Averaging to Determine Suitable Risk Measure Estimates

    by Tatjana Miljkovic, Bettina Grün

    Recent research in loss modeling resulted in a growing number of classes of statistical models as well as additional models being proposed within each class. Empirical results indicate that a range of models within or between model classes perform similarly well, as measured by goodness-of-fit or information criteria, when fitted to the same data set. This leads to model uncertainty and makes model selection a challenging task. This problem is particularly virulent if the resulting risk measures vary greatly between and within the model classes. We propose an approach to estimate risk measures that accounts for model selection uncertainty based on model averaging. We exemplify the application of the approach considering the class of composite models. This application considers 196 different left-truncated composite models previously used in the literature for loss modeling and arrives at point estimates for the risk measures that take model uncertainty into account. A simulation study highlights the benefits of this approach. The data set on Norwegian fire losses is used to illustrate the proposed methodology.