David S. Brée

Prediction accuracy and sloppiness of log-periodic functions

We show that log-periodic power-law (LPPL) functions are intrinsically very hard to fit to time series. This comes from their sloppiness, the squared residuals depending very much on some combinations of parameters and very little on other ones. The time of singularity that is supposed to give an estimate of the day of the crash belongs to the latter category. We discuss in detail why and how the fitting procedure must take into account the sloppy nature of this kind of model. We then test the reliability of LPPLs on synthetic AR(1) data replicating the Hang Seng 1987 crash and show that even this case is borderline regarding predictability of divergence time. We finally argue that current methods used to estimate a probabilistic time window for the divergence time are likely to be over-optimistic.

Stochastic volatility and the goodness-of-fit of the Heston model

Recently, Drăgulescu and Yakovenko proposed an analytical formula for computing the probability density function of stock log returns, based on the Heston model, which they tested empirically. Their research design inadvertently favourably biased the fit of the data to the Heston model, thus overstating their empirical results. Furthermore, Drăgulescu and Yakovenko did not perform any goodness-of-fit statistical tests. This study employs a research design that facilitates statistical tests of the goodness-of-fit of the Heston model to empirical returns. Robustness checks are also performed. In brief, the Heston model outperformed the Gaussian model only at high frequencies and even so does not provide a statistically acceptable fit to the data. The Gaussian model performed (marginally) better at medium and low frequencies, at which points the extra parameters of the Heston model have adverse impacts on the test statistics.

A clock-less implementation of the AES resists to power and timing attacks

New cryptanalytical techniques, in particular, power and timing analysis, pose a serious threat to cryptographic devices such as smart cards. By analyzing the power dissipation or timing of encryptions in a device, encrypted information inside can be deduced. The weakness is not in the encryption algorithms themselves, but in their implementations. We show that not even the new advanced encryption standard (AES), when implemented in conventional hardware, is secure from power attacks; a few power samples were enough to deduce the secret key. A new specially designed implementation of the AES on a clock-less dual-rail chip is presented and shown to possess a very considerable improvement against power attacks compared to the conventional design. This implementation is also resistant to timing, fault induction and clock glitch attacks.

Towards a Formalization of the Semantics of Some Temporal Prepositions

Many prepositions can be used to convey temporal information relevant to the period of a proposition. Three types of information are identified: the duration of the proposition (floating), its duration relative to the time of reference of the discourse and its absolute location on the time axis. Some prepositions can be used in all three manners, others in only one; rules are given for determining in which manner the former are being used in any given sentence. Within each manner, there are several different possible uses to which temporal prepositions are put. These uses distinguish between existential and universal quantification over time, indicate whether or not the extremes of the period are to be included and mark for motion through time. To capture these a semantics for the 9 temporal prepositions indicating periods is provided using the first-order predicate calculus. An explanation is proffered for why some of these uses can be filled by two prepositions.

The Effect of Numeric Features on the Scalability of Inductive Learning Programs

The behaviour of a learning program as the quantity of data increases affects to a large extent its applicability on real-world problems. This paper presents the results of a theoretical and experimental investigation of the scalability of four well-known empirical concept learning programs. In particular it examines the effect of using numeric features in the training set. The theoretical part of the work involved a detailed worst-case computational complexity analysis of the algorithms. The results of the analysis deviate substantially from previously reported estimates, which have mainly examined discrete and finite feature spaces. In order to test these results, a set of experiments was carried out, involving one artificial and two real data sets. The artificial data set introduces a near-worst-case situation for the examined algorithms, while the real data sets provide an indication of their average-case behaviour.

Dynamical Social Psychology: An Introduction

In this chapter we outline the background to dynamical social psychology as it stood before the research described in later chapters of this book. This background will help readers who are not familiar with either social psychology or complexity science to follow those chapters more easily. It will focus on two domains within dynamical social psychology: social influence on opinion formation and the concept of self. It will also consider three aspects of dynamical social psychology that set it apart from previous theories of social psychology: the effect of the degree of coherence between the elements of a system, which explains the different behaviors we see under different circumstances, how emotions regulate other psychological systems, and the drive to minimalism, by which behavior which appears to be complex may be understood from a model of the underlying elements interacting under simple rules.

Unusual Spatial Patterns of Industrial Firm Locations Uncover their Social Interactions

Keywords-business networks; communication networks; social interaction; spatial statistics; stationary point processes; economy; industrial firms; Abstract—In this paper we report evidence from the Italian industrial sectors whereby firms that buy and sell are spatially distributed with a pattern that reflects the microeconomic powers at play. The main finding is that firms are neither clustered around population centers nor are they situated at random. Although geography has an important role in shaping the population map of Italy, the reasons for the positional pattern of buyers and sellers appear to be social. Geographic proximity between sellers and their buyers is supported by the excess in short-distance social ties.

Significance of Novel WSD Algorithms

Abstract Word sense disambiguation (WSD) is considered to be a difficult problem in computational linguistics. Single-approach solutions to this problem consisting of only one module are unlikely to yield high performance levels, while hybrid systems formed by combining a number of modules tend to offer a better performance in tackling WSD problems. We propose the strong POS + Frequency baseline as a basic easy-to-implement platform for testing how well algorithms can do when combined with other high-accuracy modules. After giving an overview of the field, we discuss our proposed novel model for WSD, which is a stand-alone contribution in a field in which ideas are repeated. Under the umbrella of our novel WSD model, called Sense Space Model (SSM), we show that significant and interesting algorithms exist. While the accuracy of some of the unsupervised offspring algorithms of the model can be low compared to the strong POS + Frequency baseline (and also compared to the top hybrid systems), sometimes even having an accuracy lower than a random system, such algorithms can still act significantly better than a random system when combined to the strong baseline, considering a meticulous 1% significance level. Therefore, ruling out such lower-accuracy modules from a hybrid system, which might otherwise appear to be a necessary elimination, is challenged. One of these significant algorithms was recently improved by introducing “a threshold” and could beat the implemented POS + Frequency baseline. This confirms that considering such lower-accuracy algorithms as significant is reasonable.

Using non-parametric search algorithms to forecast daily excess stock returns.

Are the learning procedures of genetic algorithms (GAs) able to generate optimal architectures for artificial neural networks (ANNs) in high frequency data? In this experimental study, GAs are used to identify the best architecture for ANNs. Additional learning is undertaken by the ANNs to forecast daily excess stock returns. No ANN architectures were able to outperform a random walk, despite the finding of non-linearity in the excess returns. This failure is attributed to the absence of suitable ANN structures and further implies that researchers need to be cautious when making inferences from ANN results that use high frequency data.