A conditional random field-based downscaling method for assessment of climate change impact on multisite daily precipitation in the Mahanadi basin

Downscaling to station-scale hydrologic variables from large-scale atmospheric variables simulated by general circulation models (GCMs) is usually necessary to assess the hydrologic impact of climate change. This work presents CRF-downscaling, a new probabilistic downscaling method that represents the daily precipitation sequence as a conditional random field (CRF). The conditional distribution of the precipitation sequence at a site, given the daily atmospheric (large-scale) variable sequence, is modeled as a linear chain CRF. CRFs do not make assumptions on independence of observations, which gives them flexibility in using high-dimensional feature vectors. Maximum likelihood parameter estimation for the model is performed using limited memory Broyden-Fletcher-Goldfarb-Shanno (L-BFGS) optimization. Maximum a posteriori estimation is used to determine the most likely precipitation sequence for a given set of atmospheric input variables using the Viterbi algorithm. Direct classification of dry/wet days as well as precipitation amount is achieved within a single modeling framework. The model is used to project the future cumulative distribution function of precipitation. Uncertainty in precipitation prediction is addressed through a modified Viterbi algorithm that predicts the n most likely sequences. The model is applied for downscaling monsoon (June-September) daily precipitation at eight sites in the Mahanadi basin in Orissa, India, using the MIROC3.2 medium-resolution GCM. The predicted distributions at all sites show an increase in the number of wet days, and also an increase in wet day precipitation amounts. A comparison of current and future predicted probability density functions for daily precipitation shows a change in shape of the density function with decreasing probability of lower precipitation and increasing probability of higher precipitation.

Conditional Probability-Based Significance Tests for Sequential Patterns in Multineuronal Spike Trains

We consider the problem of detecting statistically significant sequential patterns in multineuronal spike trains. These patterns are characterized by ordered sequences of spikes from different neurons with specific delays between spikes. We have previously proposed a data-mining scheme to efficiently discover such patterns, which occur often enough in the data. Here we propose a method to determine the statistical significance of such repeating patterns. The novelty of our approach is that we use a compound null hypothesis that not only includes models of independent neurons but also models where neurons have weak dependencies. The strength of interaction among the neurons is represented in terms of certain pair-wise conditional probabilities. We specify our null hypothesis by putting an upper bound on all such conditional probabilities. We construct a probabilistic model that captures the counting process and use this to derive a test of significance for rejecting such a compound null hypothesis. The structure of our null hypothesis also allows us to rank-order different significant patterns. We illustrate the effectiveness of our approach using spike trains generated with a simulator.

Testing against changing correlation

A test for time-varying correlation is developed within the framework of a dynamic conditional score (DCS) model for both Gaussian and Student t-distributions. The test may be interpreted as a Lagrange multiplier test and modified to allow for the estimation of models for time-varying volatility in the individual series. Unlike standard moment-based tests, the score-based test statistic includes information on the level of correlation under the null hypothesis and local power arguments indicate the benefits of doing so. A simulation study shows that the performance of the score-based test is strong relative to existing tests across a range of data generating processes. An application to the Hong Kong and South Korean equity markets shows that the new test reveals changes in correlation that are not detected by the standard moment-based test.

On measurements of formation, growth, hygroscopicity, and volatility of atmospheric ultrafine particles

Atmospheric aerosol particles affect the global climate as well as human health. In this thesis, formation of nanometer sized atmospheric aerosol particles and their subsequent growth was observed to occur all around the world. Typical formation rate of 3 nm particles at varied from 0.01 to 10 cm-3s-1. One order of magnitude higher formation rates were detected in urban environment. Highest formation rates up to 105 cm-3s-1 were detected in coastal areas and in industrial pollution plumes. Subsequent growth rates varied from 0.01 to 20 nm h-1. Smallest growth rates were observed in polar areas and the largest in the polluted urban environment. This was probably due to competition between growth by condensation and loss by coagulation. Observed growth rates were used in the calculation of a proxy condensable vapour concentration and its source rate in vastly different environments from pristine Antarctica to polluted India. Estimated concentrations varied only 2 orders of magnitude, but the source rates for the vapours varied up to 4 orders of magnitude. Highest source rates were in New Delhi and lowest were in the Antarctica. Indirect methods were applied to study the growth of freshly formed particles in the atmosphere. Also a newly developed Water Condensation Particle Counter, TSI 3785, was found to be a potential candidate to detect water solubility and thus indirectly composition of atmospheric ultra-fine particles. Based on indirect methods, the relative roles of sulphuric acid, non-volatile material and coagulation were investigated in rural Melpitz, Germany. Condensation of non-volatile material explained 20-40% and sulphuric acid the most of the remaining growth up to a point, when nucleation mode reached 10 to 20 nm in diameter. Coagulation contributed typically less than 5%. Furthermore, hygroscopicity measurements were applied to detect the contribution of water soluble and insoluble components in Athens. During more polluted days, the water soluble components contributed more to the growth. During less anthropogenic influence, non-soluble compounds explained a larger fraction of the growth. In addition, long range transport to a measurement station in Finland in a relatively polluted air mass was found to affect the hygroscopicity of the particles. This aging could have implications to cloud formation far away from the pollution sources.

Deeper and wider fully convolutional network coupled with conditional random fields for scene labeling

Deep convolutional neural networks (DCNNs) have been employed in many computer vision tasks with great success due to their robustness in feature learning. One of the advantages of DCNNs is their representation robustness to object locations, which is useful for object recognition tasks. However, this also discards spatial information, which is useful when dealing with topological information of the image (e.g. scene labeling, face recognition). In this paper, we propose a deeper and wider network architecture to tackle the scene labeling task. The depth is achieved by incorporating predictions from multiple early layers of the DCNN. The width is achieved by combining multiple outputs of the network. We then further refine the parsing task by adopting graphical models (GMs) as a post-processing step to incorporate spatial and contextual information into the network. The new strategy for a deeper, wider convolutional network coupled with graphical models has shown promising results on the PASCAL-Context dataset.

Sequential split vector quantization of LSF parameters using conditional PDF

A better performing product code vector quantization (VQ) method is proposed for coding the line spectrum frequency (LSF) parameters; the method is referred to as sequential split vector quantization (SeSVQ). The split sub-vectors of the full LSF vector are quantized in sequence and thus uses conditional distribution derived from the previous quantized sub-vectors. Unlike the traditional split vector quantization (SVQ) method, SeSVQ exploits the inter sub-vector correlation and thus provides improved rate-distortion performance, but at the expense of higher memory. We investigate the quantization performance of SeSVQ over traditional SVQ and transform domain split VQ (TrSVQ) methods. Compared to SVQ, SeSVQ saves 1 bit and nearly 3 bits, for telephone-band and wide-band speech coding applications respectively.

Three Essays on the Volatility of Finnish Stock Returns (summary section only)

First, in Essay 1, we test whether it is possible to forecast Finnish Options Index return volatility by examining the out-of-sample predictive ability of several common volatility models with alternative well-known methods; and find additional evidence for the predictability of volatility and for the superiority of the more complicated models over the simpler ones. Secondly, in Essay 2, the aggregated volatility of stocks listed on the Helsinki Stock Exchange is decomposed into a market, industry-and firm-level component, and it is found that firm-level (i.e., idiosyncratic) volatility has increased in time, is more substantial than the two former, predicts GDP growth, moves countercyclically and as well as the other components is persistent. Thirdly, in Essay 3, we are among the first in the literature to seek for firm-specific determinants of idiosyncratic volatility in a multivariate setting, and find for the cross-section of stocks listed on the Helsinki Stock Exchange that industrial focus, trading volume, and block ownership, are positively associated with idiosyncratic volatility estimates––obtained from both the CAPM and the Fama and French three-factor model with local and international benchmark portfolios––whereas a negative relation holds between firm age as well as size and idiosyncratic volatility.

Modeling Nonlinearities and Asymmetries in Asset Pricing

Financial time series tend to behave in a manner that is not directly drawn from a normal distribution. Asymmetries and nonlinearities are usually seen and these characteristics need to be taken into account. To make forecasts and predictions of future return and risk is rather complicated. The existing models for predicting risk are of help to a certain degree, but the complexity in financial time series data makes it difficult. The introduction of nonlinearities and asymmetries for the purpose of better models and forecasts regarding both mean and variance is supported by the essays in this dissertation. Linear and nonlinear models are consequently introduced in this dissertation. The advantages of nonlinear models are that they can take into account asymmetries. Asymmetric patterns usually mean that large negative returns appear more often than positive returns of the same magnitude. This goes hand in hand with the fact that negative returns are associated with higher risk than in the case where positive returns of the same magnitude are observed. The reason why these models are of high importance lies in the ability to make the best possible estimations and predictions of future returns and for predicting risk.

Maximum Loss Calculation using Scenario Analysis, Heavy Tails and Implied Volatility Patterns

The objective of this paper is to improve option risk monitoring by examining the information content of implied volatility and by introducing the calculation of a single-sum expected risk exposure similar to the Value-at-Risk. The figure is calculated in two steps. First, there is a need to estimate the value of a portfolio of options for a number of different market scenarios, while the second step is to summarize the information content of the estimated scenarios into a single-sum risk measure. This involves the use of probability theory and return distributions, which confronts the user with the problems of non-normality in the return distribution of the underlying asset. Here the hyperbolic distribution is used to describe one alternative for dealing with heavy tails. Results indicate that the information content of implied volatility is useful when predicting future large returns in the underlying asset. Further, the hyperbolic distribution provides a good fit to historical returns enabling a more accurate definition of statistical intervals and extreme events.

Evaluating the Predictability of the Volatility Smile Using Return Distributions

In this paper, we examine the predictability of observed volatility smiles in three major European index options markets, utilising the historical return distributions of the respective underlying assets. The analysis involves an application of the Black (1976) pricing model adjusted in accordance with the Jarrow-Rudd methodology as proposed in 1982. Thereby we adjust the expected future returns for the third and fourth central moments as these represent deviations from normality in the distributions of observed returns. Thus, they are considered one possible explanation to the existence of the smile. The obtained results indicate that the inclusion of the higher moments in the pricing model to some extent reduces the volatility smile, compared with the unadjusted Black-76 model. However, as the smile is partly a function of supply, demand, and liquidity, and as such intricate to model, this modification does not appear sufficient to fully capture the characteristics of the smile.

Unrealized expectations of jumps in volatility: An explanation to the low and time-varying predictive power of implied volatility

The low predictive power of implied volatility in forecasting the subsequently realized volatility is a well-documented empirical puzzle. As suggested by e.g. Feinstein (1989), Jackwerth and Rubinstein (1996), and Bates (1997), we test whether unrealized expectations of jumps in volatility could explain this phenomenon. Our findings show that expectations of infrequently occurring jumps in volatility are indeed priced in implied volatility. This has two important consequences. First, implied volatility is actually expected to exceed realized volatility over long periods of time only to be greatly less than realized volatility during infrequently occurring periods of very high volatility. Second, the slope coefficient in the classic forecasting regression of realized volatility on implied volatility is very sensitive to the discrepancy between ex ante expected and ex post realized jump frequencies. If the in-sample frequency of positive volatility jumps is lower than ex ante assessed by the market, the classic regression test tends to reject the hypothesis of informational efficiency even if markets are informationally effective.

Short-Horizon Asymmetric Mean-Reversion and Overreactions: Evidence from the Nordic Stock Markets

This paper examines the asymmetric behavior of conditional mean and variance. Short-horizon mean-reversion behavior in mean is modeled with an asymmetric nonlinear autoregressive model, and the variance is modeled with an Exponential GARCH in Mean model. The results of the empirical investigation of the Nordic stock markets indicates that negative returns revert faster to positive returns when positive returns generally persist longer. Asymmetry in both mean and variance can be seen on all included markets and are fairly similar. Volatility rises following negative returns more than following positive returns which is an indication of overreactions. Negative returns lead to increased variance and positive returns leads even to decreased variance.

Examining and Modeling the Dynamics of the Volatility Surface

The objective of this paper is to investigate and model the characteristics of the prevailing volatility smiles and surfaces on the DAX- and ESX-index options markets. Continuing on the trend of Implied Volatility Functions, the Standardized Log-Moneyness model is introduced and fitted to historical data. The model replaces the constant volatility parameter of the Black & Scholes pricing model with a matrix of volatilities with respect to moneyness and maturity and is tested out-of-sample. Considering the dynamics, the results show support for the hypotheses put forward in this study, implying that the smile increases in magnitude when maturity and ATM volatility decreases and that there is a negative/positive correlation between a change in the underlying asset/time to maturity and implied ATM volatility. Further, the Standardized Log-Moneyness model indicates an improvement to pricing accuracy compared to previous Implied Volatility Function models, however indicating that the parameters of the models are to be re-estimated continuously for the models to fully capture the changing dynamics of the volatility smiles.

An Empirical Study of the Mixture of Time and Movements in Prices

This paper investigates the persistent pattern in the Helsinki Exchanges. The persistent pattern is analyzed using a time and a price approach. It is hypothesized that arrival times are related to movements in prices. Thus, the arrival times are defined as durations and formulated as an Autoregressive Conditional Duration (ACD) model as in Engle and Russell (1998). The prices are defined as price changes and formulated as a GARCH process including duration measures. The research question follows from market microstructure predictions about price intensities defined as time between price changes. The microstructure theory states that long transaction durations might be associated with both no news and bad news. Accordingly, short durations would be related to high volatility and long durations to low volatility. As a result, the spread will tend to be larger under intensive moments. The main findings of this study are 1) arrival times are positively autocorrelated and 2) long durations are associated with low volatility in the market.

Intraday Linkages across International Equity Markets

Utilizing concurrent 5-minute returns, the intraday dynamics and inter-market dependencies in international equity markets were investigated. A strong intraday cyclical autocorrelation structure in the volatility process was observed to be caused by the diurnal pattern. A major rise in contemporaneous cross correlation among European stock markets was also noticed to follow the opening of the New York Stock Exchange. Furthermore, the results indicated that the returns for UK and Germany responded to each other’s innovations, both in terms of the first and second moment dependencies. In contrast to earlier research, the US stock market did not cause significant volatility spillover to the European markets.