arXiv subjects

Henryk Gzyl

Publications and source records attributed to Henryk Gzyl.

33 records · Page 2Linked to original sources

arXiv2020

How dark is the dark side of diversification?

Against the widely held belief that diversification at banking institutions contributes to the stability of the financial system, Wagner (2010) found that diversification actually makes systemic crisis more likely. While it is true, as Wagner asserts, that the probability of joint default of the diversified portfolios is larger; we contend that, as common practice, the effect of diversification is examined with respect to a risk measure like VaR. We find that when banks use VaR, diversification does reduce individual and systemic risk. This, in turn, generates a different set of incentives for banks and regulators.

q-fin.RM↗

arXiv2020

From generalized arithmetic means to geodesics to Hamilton dynamics to Bregman divergences

Here we examine some connections between the notions of generalized arithmetic means, geodesics, Lagrange-Hamilton dynamics and Bregman divergences. In a previous paper we developed a predictive interpretation of generalized arithmetic means. That work was more probabilistically oriented. Here we take a geometric turn, and see that generalized arithmetic means actually minimize a geodesic distance on $\mathbb{R}^n.$ Such metrics might result from pull-backs of the Euclidean metric in $\mathbb{R}^n.$ We shall furthermore see that in some cases these pull-backs might coincide with the Hessian of a convex function. This occurs when the Hessian of a convex function has a square root that is the Jacobian of a diffeomorphism in $\mathbb{R}^n.$ In this case we obtain a comparison between the Bregman divergence defined by the convex function and the geodesic distance in the metric defined by its Hessian.

math.OC↗

arXiv2020

A predictive approach to generalized arithmetic means

The goal of this note is to provide a geometric setting in which generalized arithmetic means are best predictors in an appropriate metric. This characterization provides a geometric interpretation to the concept of certainty equivalent. Besides that, in this geometric setting there also exists the notion of conditional expectation as best predictor given prior information. This leads to a notion of conditional preference and to the notion of conditional certainty equivalent, which turns out to be consistent with the notion of fair pricing.

math.PR↗

arXiv2018

Geometry and clustering with metrics derived from separable Bregman divergences

Separable Bregman divergences induce Riemannian metric spaces that are isometric to the Euclidean space after monotone embeddings. We investigate fixed rate quantization and its codebook Voronoi diagrams, and report on experimental performances of partition-based, hierarchical, and soft clustering algorithms with respect to these Riemann-Bregman distances.

cs.LG↗

arXiv2018

Prediction in logarithmic distance

The metric properties of the set in which random variables take their values lead to relevant probabilistic concepts. For example, the mean of a random variable is a best predictor in that it minimizes the standard Euclidean distance or $L_2$ norm in an appropriate class of random variables. Similarly, the median is the same concept but when the distance is measured by the $L_1$ norm. These two predictors stem from the fact that the mean and the median, minimize the distance to a given set of points when distances in $\mathbb{R}$ or in $\mathbb{R}^n$ are measured in the aforementioned metrics.\\ It so happens that an interesting {\it logarithmic distance} can be defined on the cone of strictly positive vectors in $\mathbb{R}^n$ in such a way that the minimizer of the distance to a collection of points is their geometric mean.\\ This distance on the base space leads to an interesting distance on the class of strictly positive random variables, which in turn leads to an interesting class of best predictors predictors and their estimators, as well as a corresponding notion of conditional expectation. The appropriate version of the Law of Large Numbers and the Central Limit Theorem, can also be obtained. We shall see that, for example, the lognormal variables are the analogue of the Gaussian variables for the modified version of the Central Limit Theorem.

math.PR↗

arXiv2018

Prediction in Riemannian metrics derived from divergence functions

Divergence functions are interesting discrepancy measures. Even though they are not true distances, we can use them to measure how separated two points are. Curiously enough, when they are applied to random variables, they lead to a notion of best predictor that coincides with usual best predictor in Euclidean distance. Given a divergence function, we can derive from it a Riemannian metric, which leads to a distance in which means and best predictors do not coincide with their Euclidean counterparts. It is the purpose of this note to study the Riemannian metric derived from the divergence function as well as its use in prediction theory.

math.ST↗

arXiv2016

Superresolution in the maximum entropy approach to invert Laplace transforms

The method of maximum entropy has proven to be a rather powerful way to solve the inverse problem consisting of determining a probability density $f_S(s)$ on $[0,\infty)$ from the knowledge of the expected value of a few generalized moments, that is, of functions $g_i(S)$ of the variable $S.$ A version of this problem, of utmost relevance for banking, insurance, engineering and the physical sciences, corresponds to the case in which $S \geq 0$ and $g_i(s)=\exp(-α_i s),$ th expected values $E[\exp-α_i S)]$ are the values of the Laplace transform of $S$ the points $α_i$ on the real line. Since inverting the Laplace transform is an ill-posed problem, to devise numerical tecniques that are efficient is of importance for many applications, specially in cases where all we know is the value of the transform at a few points along the real axis. A simple change of variables transforms the Laplace inversion problem into a fractional moment problem on $[0,1].$ It is remarkable that the maximum entropy procedure allows us to determine the density on $[0,1]$ with high accuracy. In this note, we examine why this might be so.

math.OC↗

arXiv2015

Sample dependence in the maximum entropy solution to the generalized moment problem

The method of maximum entropy is quite a powerful tool to solve the generalized moment problem, which consists of determining the probability density of a random variable X from the knowledge of the expected values of a few functions of the variable. In actual practice, such expected values are determined from empirical samples, leaving open the question of the dependence of the solution upon the sample. It is the purpose of this note to take a few steps towards the analysis of such dependence.

math.ST↗

arXiv2014

Two maxentropic approaches to determine the probability density of compound risk losses

Here we present an application of two maxentropic procedures to determine the probability density distribution of compound sums of random variables, using only a finite number of empirically determined fractional moments. The two methods are the Standard method of Maximum Entropy (SME), and the method of Maximum Entropy in the Mean (MEM). We shall verify that the reconstructions obtained satisfy a variety of statistical quality criteria, and provide good estimations of VaR and TVaR, which are important measures for risk management purposes. We analyze the performance and robustness of these two procedures in several numerical examples, in which the frequency of losses is Poisson and the individual losses are lognormal random variables. As side product of the work, we obtain a rather accurate description of the density of the compound random variable. This is an extension of a previous application based on the Standard Maximum Entropy approach (SME) where the analytic form of the Laplace transform was available to a case in which only observed or simulated data is used. These approaches are also used to develop a procedure to determine the distribution of the individual losses through the knowledge of the total loss. Then, in the case of having only historical total losses, it is possible to decompound or disaggregate the random sums in its frequency/severity distributions, through a probabilistic inverse problem.

q-fin.RM↗

arXiv2009

Predictive approach to some quantum paradoxes

In classical probability theory, the best predictor of a future observation of a random variable $X,$ is its expected value $E_P[X]$ when no other information is available When information consisting in the observation of another random variable $Y$ is available, then the best predictor of $X$ is another random variable $E_P[X|Y].$ It is the purpose of this note to explore the analogue of this in the case of quantum mechanics. We shall see that exactly as in classical prediction theory, when the result of an observation is taken into account by means of a non-commutative conditional expectation, some of the usual paradoxes cease to be such.

math-ph↗

arXiv2009

Stochastic Volatility Models Including Open, Close, High and Low Prices

Mounting empirical evidence suggests that the observed extreme prices within a trading period can provide valuable information about the volatility of the process within that period. In this paper we define a class of stochastic volatility models that uses opening and closing prices along with the minimum and maximum prices within a trading period to infer the dynamics underlying the volatility process of asset prices and compares it with similar models that have been previously presented in the literature. The paper also discusses sequential Monte Carlo algorithms to fit this class of models and illustrates its features using both a simulation study and data form the SP500 index.

q-fin.ST↗

arXiv2007

Filtering Additive Measurement Noise with Maximum Entropy in the Mean

The purpose of this note is to show how the method of maximum entropy in the mean (MEM) may be used to improve parametric estimation when the measurements are corrupted by large level of noise. The method is developed in the context on a concrete example: that of estimation of the parameter in an exponential distribution. We compare the performance of our method with the bayesian and maximum likelihood approaches.

cs.LG↗

arXiv2007

Assessment and Propagation of Input Uncertainty in Tree-based Option Pricing Models

This paper aims to provide a practical example on the assessment and propagation of input uncertainty for option pricing when using tree-based methods. Input uncertainty is propagated into output uncertainty, reflecting that option prices are as unknown as the inputs they are based on. Option pricing formulas are tools whose validity is conditional not only on how close the model represents reality, but also on the quality of the inputs they use, and those inputs are usually not observable. We provide three alternative frameworks to calibrate option pricing tree models, propagating parameter uncertainty into the resulting option prices. We finally compare our methods with classical calibration-based results assuming that there is no options market established. These methods can be applied to pricing of instruments for which there is not an options market, as well as a methodological tool to account for parameter and model uncertainty in theoretical option pricing.

cs.CE↗

arXiv2006

Inverse problems for random walks on trees: network tomography

Let $G$ be a finite tree with root $r$ and associate to the internal vertices of $G$ a collection of transition probabilities for a simple nondegenerate Markov chain. Embedd $G$ into a graph $G^\prime$ constructed by gluing finite linear chains of length at least 2 to the terminal vertices of $G.$ Then $G^\prime$ admits distinguished boundary layers and the transition probabilities associated to the internal vertices of $G$ can be augmented to define a simple nondegenerate Markov chain $X$ on the vertices of $G^\prime.$ We show that the transition probabilities of $X$ can be recovered from the joint distribution of first hitting time and first hitting place of $X$ started at the root $r$ for the distinguished boundary layers of $G^\prime.$

math.PR↗

arXiv2006

Towards a Bayesian framework for option pricing

In this paper, we describe a general method for constructing the posterior distribution of an option price. Our framework takes as inputs the prior distributions of the parameters of the stochastic process followed by the underlying, as well as the likelihood function implied by the observed price history for the underlying. Our work extends that of Karolyi (1993) and Darsinos and Satchell (2001), but with the crucial difference that the likelihood function we use for inference is that which is directly implied by the underlying, rather than imposed in an ad hoc manner via the introduction of a function representing "measurement error." As such, an important problem still relevant for our method is that of model risk, and we address this issue by describing how to perform a Bayesian averaging of parameter inferences based on the different models considered using our framework.

cs.CE↗