Machine learning for financial transaction classification across companies using character-level word embeddings of text fields

An important initial step in accounting is mapping financial transfers to the corresponding accounts. We devised machine-learning-based systems that automate this process. They use word embeddings with character-level features to process transaction texts. When considering 473 companies independently, our approach achieved an average top-1 accuracy of 80.50%, outperforming baselines that exclude the transaction texts or rely on a lexical bag-of-words text representation. We extended the approach to generalizes across companies and even across different corporate sectors. After standardization of the account structures and careful feature engineering, a single classifier trained on 44 companies from 28 sectors achieved a test accuracy of more than 80%. When trained on 43 companies and tested on the remaining one, the system achieved an average performance of 64.62%. This rate increased to nearly 70% when considering only the largest sector.     

The Multilevel Approach to Repeated Measures for Complete and Incomplete Data

Repeated measurements often are analyzed by multivariate analysis of variance (MANOVA). An alternative approach is provided by multilevel analysis, also called the hierarchical linear model (HLM), which makes use of random coefficient models. This paper is a tutorial which indicates that the HLM can be specified in many different ways, corresponding to different sets of assumptions about the covariance matrix of the repeated measurements. The possible assumptions range from the very restrictive compound symmetry model to the unrestricted multivariate model. Thus, the HLM can be used to steer a useful middle road between the two traditional methods for analyzing repeated measurements. Another important advantage of the multilevel approach to analyzing repeated measures is the fact that it can be easily used also if the data are incomplete. Thus it provides a way to achieve a fully multivariate analysis of repeated measures with incomplete data. This revised version was published online in June 2006 with corrections to the Cover Date.     

Environmental Valuation with Imperfect Information The Case of the Random Utility Model

This paper considers welfare analysis with therandom utility model (RUM) when perceptions ofenvironmental quality differ from objectivemeasures of environmental quality. Environmental quality is assumed to be anexperience good, so that while perceptions ofquality determine choices, ex postutility is determined by objective quality. Given this assumption, I derive a measure ofthe welfare impact of changes in environmentalquality, and I show how this new welfaremeasure differs from the traditional welfaremeasure developed by Hanemann (1982). This newwelfare measure provides an approach tomeasuring the value of information aboutenvironmental quality within the framework ofthe random utility model.     

The Random Walk Hypothesis in the Emerging Indian Stock Market

This paper examines the random walk hypothesis in the emerging Indian stock market using daily data on individual stocks. The statistical evidence in this paper rejects the random walk hypothesis. The results suggest that daily returns earned by individual stocks and by an equally weighted portfolio show significant non–linear dependence and persistent volatility effects. The non–linear dependence takes the form of ARCH–type conditional heteroskedasticity and does not appear to be caused by nonstationarity of underlying economic variables. Though conditional volatility is time varying, it does not explain expected returns.     

EMBEDDED MATRICES FOR FINITE MARKOV CHAINS

For an arbitrary subset A of the finite state space 5 of a Markov chain the so–called embedded matrix P _A is introduced. By use of these matrices formulas expressing all kinds of probabilities can be written down almost automatically, and calculated very easily on a computer. Also derivations can be given very systematically.     

On efficiency of the ratio estimator

Summary The exact mean square error for the ratio estimator of a finite population total based on simple random sampling without replacement is shown to have an expected value less than that of the variance of the ratio estimator based on Midzuno’s scheme, under a usual super-population model.     

Sample Design and Allocation for Random Digit Dialling

Sample design and sample allocation methods are developed for random digit dialling in household telephone surveys. The proposed method is based on a two-way stratification of telephone numbers. A weighted probability proportional to size sample allocation technique is used, with auxiliary variables about the telephone coverage rates, within local telephone exchanges of each substrata. This makes the sampling design nearly “self-weighting” in residential numbers when the prior information is well assigned. A computer program generates random numbers for the local areas within the existing phone capacities. A simulation study has shown greater sample allocation gain by the weighted probabilities proportional to size measures over other sample allocation methods. The amount of dialling required to obtain the sample is less than for proportional allocation. A decrease is also observed on the gain in sample allocation for some methods through the increasing sample sizes.     

Probability theory in fuzzy sample spaces

This paper tries to develop a neat and comprehensive probability theory for sample spaces where the events are fuzzy subsets of The investigations are focussed on the discussion how to equip those sample spaces with suitable -algebras and metrics. In the end we can point out a unified concept of random elements in the sample spaces under consideration which is linked with compatible metrics to express random errors. The result is supported by presenting a strong law of large numbers, a central limit theorem and a Glivenko-Cantelli theorem for these kinds of random elements, formulated simultaneously w.r.t. the selected metrics. As a by-product the line of reasoning, which is followed within the paper, enables us to generalize as well as to bring together already known results and concepts from literature.Acknowledgement. The author would like to thank the participants of the 23rd Linz Seminar on Fuzzy Set Theory for the intensive discussion of the paper. Especially he is indebted to Professors Diamond and Höhle whose remarks have helped to get deeper insights into the subject. Additionally, the author is grateful to one anonymous referee for careful reading and valuable proposals which have led to an improvement of the first draft.This paper was presented at the 23rd Linz Seminar on Fuzzy Set Theory, Linz, Austria, February 5–9, 2002.     

Assessment of Interjudge Reliability in the Open-Ended Questions Coding Process

In the process of coding open-ended questions, the evaluation of interjudge reliability is a critical issue. In this paper, using real data, the behavior of three coefficients of reliability among coders, Cohen’s K, Krippendorff’s α and Perreault and Leigh’s I _r are patterned, in terms of the number of judges involved and the categories of answer defined. The outcome underlines the importance of both variables in the valuations of interjudge reliability, as well as the higher adequacy of Perreault and Leigh’s I _r and Krippendorff’s α for marketing and opinion research.