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.     

Integrated bio-economic models as tools to support land-use decision making: a review of potential and limitations

Bio-economic modelling has become a useful tool for anticipating the outcomes of policies and technologies before their implementation. Advances in mathematical programming have made it possible to build more comprehensive models. In an overview of recent studies about bio-economic models applied to land-use problems in agriculture and forestry, we evaluated how aspects such as uncertainty, multiple objective functions, system dynamics and time have been incorporated into models. We found that single objective models were more frequently applied at the farm level, while multiple objective modelling has been applied to meet concerns at the landscape level. Among the objectives, social aspects are seldom represented in all models, when being compared to economic and environmental aspects. The integration of uncertainty is occasionally a topic, while stochastic approaches are more frequently applied than non-stochastic robust methods. Most multiple-objective models do not integrate uncertainty or sequential decision making. Static approaches continue to be more recurrent than truly dynamic models. Even though integrating multiple aspects may enhance our understanding of a system; it involves a tradeoff between complexity and robustness of the results obtained. Land-use models have to address this balance between complexity and robustness in order to evolve towards robust multiple-objective spatial optimization as a prerequisite to achieve sustainability goals.     

Stock market volatility forecasting: Do we need high-frequency data?

The general consensus in the volatility forecasting literature is that high-frequency volatility models outperform low-frequency volatility models. However, such a conclusion is reached when low-frequency volatility models are estimated from daily returns. Instead, we study this question considering daily, low-frequency volatility estimators based on open, high, low, and close daily prices. Our data sample consists of 18 stock market indices. We find that high-frequency volatility models tend to outperform low-frequency volatility models only for short-term forecasts. As the forecast horizon increases (up to one month), the difference in forecast accuracy becomes statistically indistinguishable for most market indices. To evaluate the practical implications of our results, we study a simple asset allocation problem. The results reveal that asset allocation based on high-frequency volatility model forecasts does not outperform asset allocation based on low-frequency volatility model forecasts.     

Do interventions in foreign exchange markets modify investors' expectations? The experience of Japan between 1992 and 2004

The purpose of this paper is to analyze the impact of the Bank of Japan's official interventions on the JPY/USD parity during the period 1992–2004. The novelty of our approach is to combine two recent advances of the empirical literature on foreign exchange interventions: (i) drawing on over-the-counter option prices to characterize more precisely the distribution of market expectations; (ii) redefining interventions in terms of events as they tend to come in clusters. Moreover, in order to deal with the features of the data (small sample size, non-standard distribution), we use bootstrap tests.We show that interventions have a significant impact on the mean expectation (the forward rate). The results are more ambiguous for variance. Additionally, we find that the effect of interventions on skewness is significant, robust to different definitions of skewness, and consistent with the direction of interventions. On the contrary, our results clearly show that kurtosis is not affected by interventions. We finally show that: (i) coordination increases effectiveness of interventions; (ii) results are not altered when controlling for other economic and political news.     

An alternative interpretation of multiple bounded WTP data—Certainty dependent payment card intervals

This paper analyzes the multiple bounded format, in which uncertainty is directly incorporated into the WTP question. A new approach for analyzing multiple bounded uncertainty data is presented. The intuition underlying the approach is that uncertain individuals would like to state their WTP as intervals rather than precise values and that the width of the intervals is determined by the degree of uncertainty. The approach is compared to the one applied in Welsh and Poe [Welsh, M., Poe, G.L., 1998. Elicitation effects in contingent valuation: comparisons to a multiple bounded discrete choice approach. Journal of Environmental Economics and Management 36, 170–185] which treats uncertainty by conditioning responses on specific verbal probability statements. We argue that the conditioning approach overestimates mean and median WTP and that conditioning WTP estimates on probability statements like “probably” and “unsure” make them “fuzzy”. To empirically compare the two approaches we use data from 2004 concerning implementation of a predator protection policy in Sweden. Our analysis show that the suggested approach: (1) is more intuitive; (2) better fits the data; (3) estimates mean and median WTP with better precision; (4) is less sensitive to distributional assumptions; and (5) it is better suited for policy analysis.     

A market model of analysts’ opinions to explain changes in the dispersion of opinions

The demand for and supply of analysts’ opinions in this model yield an equilibrium that demonstrates how the information content of the opinions reacts to changes in exogenous parameters. The model also shows how changes in the parameters make analysts’ opinions more or less dispersed; for example, a decline in investor risk aversion, a decrease in market volatility, and an increase in information costs can lead to analysts’ opinions becoming more similar. Recognizing how exogenous factors can affect the supply and demand of analysts’ opinions gives additional insights into questions concerning what may appear to be herd behavior by analysts and also the relationship between forecast dispersion and information content. (JEL: G29, C71)     

How to measure mutual fund performance: economic versus statistical relevance

In the present paper a comprehensive assessment of existing mutual fund performance models is presented. Using a survivor‐bias free database of all US mutual funds, we explore the added value of introducing extra variables such as size, book‐to‐market, momentum and a bond index. In addition to that we evaluate the use of introducing time‐variation in betas and alpha. The search for the most suitable model to measure mutual fund performance will be addressed along two lines. First, we are interested in the statistical significance of adding more factors to the single factor model. Second, we focus on the economic importance of more elaborate model specifications. The added value of the present study lies both in the step‐wise process of identifying relevant factors, and the use of a rich US mutual fund database that was recently released by the Center for Research in Security Prices.     

The LeChatelier-Samuelson principle revisited

Within the context of a linear Leontief model, the LeChatelier-Samuelson principle examines the effects of an increase in some final demand on the output levels under the constraint that the production of certain goods is held at its original value. The principle states that the increase in any output is larger when fewer output levels are kept constant. The present paper discusses bounds for such incremental changes, second-order effects, the consequences on the markets for the products with restricted output levels, and generalizations of the original assumptions.I would like to thank two anonymous referees for their helpful comments.     

Trader Anonymity, Price Formation and Liquidity

Using data from the Frankfurt Stock Exchange we analyze priceformation and liquidity in a non-anonymous environment withsimilarities to the floor of the NYSE. Our main hypothesis isthat the non-anonymity allows the specialist to assess the probabilitythat a trader trades on the basis of private information. Heuses this knowledge to price discriminate. This can be achievedby quoting a large spread and granting price improvement totraders deemed uninformed. Consistent with our hypothesis wefind that price improvement reflects lower adverse selectioncosts but does not lead to a reduction in the specialist's profit.Further, the quote adjustment following transactions at thequoted bid or ask price is more pronounced than the quote adjustmentafter transactions at prices inside the spread. Our resultsindicate that anonymity comes at the cost of higher adverseselection risk. JEL Classification: G10.     

Trader Anonymity, Price Formation and Liquidity

Using data from the Frankfurt Stock Exchange we analyze price formationand liquidity in a non-anonymous environment with similarities to thefloor of the NYSE. Our main hypothesis is that the non-anonymity allows the specialist to assess the probability that atrader trades on the basis of private information. He uses this knowledgeto price discriminate. This can be achieved by quoting a large spread and granting price improvement to traders deemed uninformed.Consistent with our hypothesis we find that price improvement reflects loweradverse selection costs but does not lead to a reduction in the specialist's profit. Further, the quote adjustmentfollowing transactions at the quoted bid or ask price is more pronounced than the quote adjustment aftertransactions at prices inside the spread. Our results indicate that anonymity comes at the cost ofhigher adverse selection risk.