Deep order flow imbalance: Extracting alpha at multiple horizons from the limit order book

We employ deep learning in forecasting high-frequency returns at multiple horizons for 115 stocks traded on Nasdaq using order book information at the most granular level. While raw order book states can be used as input to the forecasting models, we achieve state-of-the-art predictive accuracy by training simpler “off-the-shelf” artificial neural networks on stationary inputs derived from the order book. Specifically, models trained on order flow significantly outperform most models trained directly on order books. Using cross-sectional regressions, we link the forecasting performance of a long short-term memory network to stock characteristics at the market microstructure level, suggesting that “information-rich” stocks can be predicted more accurately. Finally, we demonstrate that the effective horizon of stock specific forecasts is approximately two average price changes.     

Marketing modeling reality and the realities of marketing modeling

This paper shows how analytic modeling research in the Marketing field is focused on answering questions of “How?” and “Why?” It describes the disciplines involved in analytic modeling; examines how the key criteria of parsimony and robustness help to define a good model; and discusses other goodness criteria, including appropriate use of analytic techniques, applicability of the model to institutionally rich, real-world problems, non-obvious results, generalizability, and ability to provide insight where other research techniques do not work. The paper defines and discusses key concepts in analytic models of distribution channels, including double marginalization, coordination, incentive alignment and contract design, strategic substitutability and complementarity, externalities, and principal–agent problems. Next, the paper summarizes research presented in the session on analytic models in channels at the Erin Anderson conference; and finally, the paper suggests avenues for future analytic modeling research.     

Dispelling the Myth

This paper investigates attempts to test the proposition that advertising is an important factor influencing total consumption. It surveys prior work in the area and discusses a more useful definition of the expression ‘changes in advertising cause changes in consumption’ than has been implied in most past tests of the relationship. The criteria for testing causality involve comparing the forecasting ability of a univariate consumption model with a bivariate model of consumption and advertising. These models are built using the powerful Box-Jenkins techniques of time series analysis. A study of the relationship in the United Kingdom is reported covering the period 1969 to 1980. The results do not support the hypothesis that total advertising expenditures alter aggregate consumer behaviour.     

中国男性人口死亡率动态预测的方法比较——基于Lee-Carter模型与贝叶斯分层模型的研究

选取中国国家统计局公布的全国男性分年龄人口死亡率,采用Lee-Carter模型和贝叶斯分层模型对男性人口死亡率进行动态预测,并将结果进行比较,探寻更加适合中国男性人口死亡率的预测方法。研究表明,Lee-Carter模型和贝叶斯分层模型在男性人口死亡率预测中各有优劣,长短期可搭配使用;Lee-Carter模型具有更佳的拟合优度和短期预测效果,但预测区间的范围较窄,且死亡率改善速度为常数的假设不符合人类生物规律;贝叶斯分层模型适合长期预测,预期寿命服从逻辑斯蒂分布的假设较为合理,然而通过模型生命表将预期寿命转换为年龄别死亡率,将会损失死亡率的年龄信息。     

基于LSTM深度学习的大豆期货价格预测

面对越来越复杂的金融市场环境,以传统统计学和计量学为主的时间序列预测模型在发现序列中的长期依赖关系方面存在一定局限性,而深度学习中的长短期记忆(LSTM)网络有望克服这一问题。通过构造一个多层LSTM网络价格预测模型,使用中国2007—2019年大豆期货价格数据进行了实证研究。结果显示,参数调优对LSTM网络模型预测效果有着较大影响,其中影响较大的主要参数包括迭代次数、学习率、窗口大小和网络层数等;与ARIMA模型、MLP模型、SVR模型相比,LSTM网络模型的预测结果准确性更高,在拟合优度(R-2)上分别提高了1.064%、2.147%、1.674%。LSTM网络模型在价格预测方面的良好表现,为预测大豆期货价格提供了新思路。     

The use of incremental goodness of fit indices in structural equation models in marketing research

Indices of the incremental goodness fit of structural equation models can provide useful information beyond that provided by statistical significance tests. Two such indices are described, and their advantages are illustrated with several marketing research studies. Additional discussion focuses on the issue of appropriate null hypotheses against which to test proposed models.     

Empirical modeling of section 162(a) (2) tax court decisions: Identifying the location of a tax home

One-hundred Tax Court cases concerning Section 162(a) (2) were subjected to several multivariate PROBIT and discriminant analyses to determine which factors best define the location of a “tax home.” In each case, the government disagreed with the taxpayer and contended that the “tax home” was either nonexistent or was located in elsewhere. Various sensitivity analyses were performed to test model specifications regarding linear or quadratic functions, discriminant or PROBIT models, and temporal stability. A seven-variable linear discriminant model achieved an 88% Lachenbruch U classification accuracy and exhibited high stability with regard to the three sensitivity analyses. Implications for taxpayers, practitioners, legislators, and researchers are discussed.     

Managerial evaluation of sales forecasting effectiveness: A MIMIC modeling approach

A Multiple Indicators and MultIple Causes (MIMIC) model is developed in which managerial evaluations of forecasting effectiveness are modeled as a function of different forecast performance criteria, namely, accuracy, bias, timeliness and cost. The model is estimated using data from a survey of export sales forecasting practices and several hypotheses linking the aforementioned criteria on effectiveness are tested. The findings indicate that evaluations of forecasting effectiveness are equally influenced by short-term accuracy and absence of overestimating bias, while timely delivery of the forecast to management is somewhat less important. Long-term accuracy, underestimation and timing of production of the forecast are not found to impact on effectiveness. Implications of the results for forecasting practice are considered and future research directions identified.     

Labor market differentials and the expected gain from search

Several recent empirical studies have used the residuals from estimated earnings equations as explanatory variables in models of on-the-job search and quit behavior. It has often been argued that the coefficients on these “market differential” variables are biased downward because estimated residuals overstate the “true” quasi-rent or disincentive to search for alternative employment. This argument—that observed wages include implicit payments for workplace and person-specific characteristics not fully specified in earnings models— ignores an opposing bias associated with the value of being able to reject unattractive wage offers. As a consequence, there is no unambiguous a priori relationship between estimated wage residuals and the theoretical expected gain from search. Some evidence bearing on the relative magnitudes of the two biases is discussed.     

Forecasting Advertising Expenditure

The 1980s are proving to be a decade of great uncertainty during which the possession of accurate short-term forecasting techniques can lead to better decision-making. Forecasts of advertising expenditure in total and by medium is an especially important source of information for the advertising industry—advertising agencies, media owners, advertisers—and all others concerned. Furthermore, it is particularly important to unravel the systematic from chance changes in the market as a whole. This paper attempts both to illustrate and to evaluate the ‘naive’ Box-Jenkins method of time series forecasting as related to aggregate advertising expenditures in Britain and West Germany.     

Using simultaneous latent structure models to analyze group differences: Exploratory analysis of buying style items

This article shows how methods for simultaneous latent structure analysis can be used to test for intergroup differences in unobservable characteristics. In many respects these techniques are similar to those for simultaneous factor analysis. To show how these methods can be used, we analyze data obtained from the 1979 Study of Media and Markets collected by the Simmons Market Research Bureau. Five dichotomized items pertaining to buying style attitudes were chosen as “indicators” of a latent variable (or variables): brand loyalty, willingness to experiment, degree of conformity to buying style of others, persuadibility, and style consciousness. The sample was composed of full-time employed, married, white females in the middle social class categories recognized on the survey. Two groups were compared in terms of latent structure model parameters. The first group consisted of those who work “at a career” and the second consisted of those whose work was “just a job.” Using the methods advocated here there is no significant difference between these two groups once the “brand loyalty” item is excluded. The model-search procedure takes account of within-group contributions to lack of fit. Generalizations of this procedure may be useful in exploratory analysis using the simultaneous latent structure model.     

Mismeasurement of market returns

Virtually all empirical tests of the Capital Asset Pricing Model have assumed (usually implicitly) that returns of some New York Stock Exchange index measure the returns of the “market factor” without error. However, recent theoretical developments suggest that this assumption may be inappropriate. In this effort, an obverse tack is adopted — the asset pricing model is assumed correct and attention is focused on the impacts of incorrect specification of market returns. A simple errors-in-variables econometric technique is used to reevaluate the oft-cited study of Black, Jensen, and Scholes. The conclusion is reached that incorrect measurement of the market is itself an onerous and unreconciled problem which adds marked bias to tests of asset pricing mechanisms.     

Demand forecasting based on natural computing approaches applied to the foodstuff retail segment

The purpose of this paper is to compare the accuracy of demand forecasting between two classical linear forecasting models (Autoregressive and Integrated Moving Average -ARIMA and Holt-Winter) and two nonlinear forecasting models based on natural computing approaches (Wavelets Neural Networks - WNN and Takagi-Sugeno Fuzzy System - TS), all applied to the aggregated retail sales of three groups of perishable food products from 2005 to 2013. Moreover, this paper evaluates the impact of demand forecasting accuracy on the demand satisfaction rate and on the overall economic performance of retail business operations. The most accurate model, WNN, had a demand satisfaction rate of 98.27% for Group A, 98.83% for Group B and 98.80% for Group C. WNN estimated a loss of revenue of R$1329.14 million/year with a minimum loss of 166 tons/year, which means that the results of WNN are 37.67% more efficient than the TS, 57.49% higher than the ARIMA and 76.79% higher than HW. This paper presents three main contributions: (i) it examines a question not evaluated in the literature on demand forecasting based on natural computing approaches in the foodstuff retail segment that generates better practical results, (ii) it proposes that a single forecasting model could be applied to different product groups and serves the organization as a whole with a good relationship between the cost and the benefit of the process and (iii) like previous studies, it proves that demand forecasting plays an important role and can generate a competitive advantage for the organization to be incorporated into its strategy.     

A Linear Multisector Model of “Endogenous” Growth and the Problem of Capital

In this paper a linear multisector model of “endogenous” growth with heterogeneous capital goods is elaborated. The purpose of this exercise is to show that this kind of model is exempt from the capital theory critique put forward against the conventional long-period neoclassical growth model á la Solow. This confirms previous claims that at least some of the “new” growth models are somewhat extraneous to neoclassical analysis and actually exhibit the logical structure of classical theory. In addition it is shown that the use of an intertemporal analysis to establish a correct long-period position is not necessary and that the use of the long-period method may speed up the elaboration of new scientific results.     

A Wolf in Sheep's Clothing: How Humanized,Retargeted Ads Intrude in Social Contexts

A great deal of past work has demonstrated that ads that “fit” or blend into their context evoke favorable responses because consumers perceive such ads as less intrusive. The current work posits that when aspects of the ad make persuasive intent salient—such as exposure to retargeted ads, which are generated based on consumers' past online browsing behavior—this relationship reverses such that fit between an ad and its context elicits higher perceptions of intrusiveness. Specifically, ads that present ad copy in a conversational style “fit” with a social media context on the dimension of social presence. Across three studies, we show that retargeted ads that “fit” the context on the dimension of social presence evoke higher levels of perceived intrusiveness, which, in turn, influence downstream responses such as attitudes toward the ad.     

Self-Sabotage and Durable-Goods Monopoly

We analyze a simple two-period linear demand durable-goods monopoly model with “self-sabotage.” The firm has the ability to sabotage its own production by increasing its future (period two) manufacturing costs. We find that an uncommitted monopoly seller has an incentive to engage in such self-sabotage, while a committed seller or renter has no such incentive. Unlike the previous papers on self-sabotage, we show this occurs even though the firm faces no rivals in the output market. In our durable-goods setting, the incentive for self-sabotage arises from the seller’s commitment problem with period-one buyers (the so-called Coase conjecture). Interestingly, we also find that this sort of self-sabotage can not only be profit enhancing for the uncommitted firm, but may also increase social welfare (in contrast to the earlier models on self-sabotage.)     

Speculative behavior of foreign exchange rates during the current float

The current regime of floating exchange rates has been characterized by a number of informed observers as economically unsatisfactory. Use of terms such as “overshooting”, “bandwagon effects”, “destabilizing”, and “insufficient speculation” reflects serious misgivings on the part of many toward the long-run viability of a floating, rather than a fixed or semi-fixed, rate regime. Using fairly standard procedures, the authors have attempted to determine the extent to which the foreign exchange market exhibits the adverse features noted above. The authors conclude that by and large foreign exchange markets have not performed particularly poorly. The foreign exchange markets seem to be efficient at least in the weak form sense. Past exchange rate changes are not useful in predicting future exchange rate changes. This empirical finding contrasts sharply with the view that the markets “overshoot”, or that there are “bandwagon effects”, or that the amount of price stabilizing speculation is inadequate.     

A two-speed Europe: How (un)stable would such a solution be?

Assuming the convergency criteria are not watered down, European Monetary Union will begin on 1.1.1999 with just a few core countries. With the help of the theory of clubs, Professor F. L. Sell analyses the incentive effects of the “monetary club” on founding members and “outsiders”.     

Should retailers encourage couples to shop together?

Many purchase decisions take place in social relationships, and yet few studies have specifically investigated couples’ purchase decisions made during shopping about products for later joint consumption. We hypothesize that romantic partners purchase more when they shop together than individually and that this effect is strong for vice products, particularly those without an organic label. For our empirical study, we asked romantic partners shopping together in a real-life context to make purchase decisions together or individually (our main experimental condition) in a self-programmed web store that offered 88 product variants (differing in category [vice/virtue] and labeling [with/without organic label]). Participants then filled out an online questionnaire on site. Results of a sequence of nested generalized linear models show that making purchase decisions together increases purchase amount (number of items selected) and purchase value (quantities selected multiplied by the corresponding willingness to pay), especially for vice products without organic labeling. In a second study, we benchmark these effects by comparing them with the effects of individual decision making and varying consumption mode (joint vs. individual consumption), using data from an online survey that followed the same structure as the main study. These effects, again estimated through generalized linear models, are negligible. Our findings strongly support the “accomplice” (rather than the “minder”) role of romantic partners in shopping. Therefore, retailers should target couples, encourage them to shop together, and emphasize joint consumption as a shopping goal.     

Single vs simultaneous equation models in capital asset pricing: The role of firm-related variables

An alternative modeling system that more adequately describes the returns-generating process than the usual single equation CAPM is provided in this paper. Firm-related variables are introduced directly into the model using a simultaneous equation system while avoiding the estimation problems of previous attempts. While not empirically testing this system with the single-equation “multi-beta” models, the relationships between the alternative processes are explored.