Mortgage default has been characterized as ruthless -or predominantly driven by the relation of house price to mortgage value -by proponents of the pure option-theoretic model and as non-ruthless by researchers who argue that transaction costs and other idiosyncratic factors determine when mortgage holders default. This work uses Fannie Mae loan performance data to present evidence supporting the hypothesis that a significant number of mortgage defaults are non-ruthless. It uses a new approach to track mortgages that are likely to default by tracking 90-day delinquent mortgages and studying which ones eventually default. It evaluates the joint put-call option embedded in a mortgage contract using a Monte Carlo simulation for the underlying stochastic variables. It identifies key differences between ruthless and non-ruthless mortgage defaults and illustrates the propensity of non-ruthless mortgage defaulters to become current on their 90-day delinquent mortgages. These observations provide valuable insights for policymakers and creditors in their task of structuring debt-relief programs for delinquent mortgage holders. It augments the analysis of mortgage defaults by considering the impact of loan-to-value ratio at mortgage origination.

No abstract

Technical analysis is frequently dismissed as an exercise in data snooping by practitioners of quantitative finance because of its lack of rigorous theoretical underpinnings. An empirical evaluation of the effectiveness of technical analysis is confounded by the subjectivity involved in identifying patterns – a higher degree of smoothing in one region may reveal patterns that were previously obscured by local noise, or it could obscure an uncovered pattern that was waiting to be identified by a technical analyst. In a seminal paper on empirical assessment of technical analysis, Lo, Mamaysky, and Wang use kernel smoothing on price charts to represent a smoothed price plot before applying geometric rules based on the order of local extrema to identify patterns used in technical analysis. The authors acknowledge the shortcomings of the procedure: constant smoothing parameter for entire price history, prescription of smoothing parameter after visual inspection of smoothed plots, and shortcomings inherent in the use of kernal smoothing‐like edge effects. Because their procedure identifies patterns based on the order of local maxima, a constant smoothing parameter is the biggest shortcoming of their algorithm – a different smoothing pattern in one region may reveal patterns that could not be discerned using existing smoothing parameters. This paper presents a more robust framework of pattern identification using probabilistic neural networks (PNNs). PNNs were chosen over other variants of neural networks – like back‐propagation‐based multi‐layer neural networks – because of the shorter training period and the relative simplicity involved in network construction for PNNs. A set of well‐known patterns is selected, and a PNN is used to identify the technical patterns observed in the price history for 30 Dow Jones Industrial components and for the S&P 500 index. Statistical tests are performed to gauge the empirical significance of profits derived from trading based on the respective technical patterns. A trading rule is added to supplement the pattern identification process and statistical tests are used to assess its profitability.

Value‐at‐risk is a standard risk metric calculated to assess the upper limit on losses incurred by a portfolio due to adverse market moves for a specified confidence level. Usually it is calculated over a 10 day period using a confidence level of 99% (95% is also common). There are three commonly used methodologies for calculating VaR (Bohdalová, 2007). These are the delta‐normal method, historical simulation and Monte‐Carlo simulation based method. Of these, Monte‐Carlo simulation based method is the most flexible because it can work with a specified probability distribution of asset returns. This work uses the probability distribution of asset prices extracted from option prices to get the VaR of a portfolio using Monte‐Carlo method.

No abstract