ArticlePDF Available


We develop and test an optimization model for maximizing response rates for online marketing research survey panels. The model consists of (1) a decision tree predictive model that classifies panelists into "states" and forecasts the response rate for panelists in each state and (2) a linear program that specifies how many panelists should be solicited from each state to maximize response rate. The model is forward looking in that it optimizes over a finite horizon during which S studies are to be fielded. It takes into account the desired number of responses for each study, the likely migration pattern of panelists between states as they are invited and respond or do not respond, as well as demographic requirements. The model is implemented using a rolling horizon whereby the optimal solution for S successive studies is derived and implemented for the first study. Then, as results are observed, an optimal solution is derived for the next S studies, and the solution is implemented for the first of these studies, etc. The procedure is field tested and shown to increase response rates significantly compared to the heuristic currently being used by panel management. Further analysis suggests that the improvement was due to the predictive model and that a "greedy algorithm" would have done equally well in the field test. However, further Monte Carlo simulations suggest circumstances under which the model would outperform the greedy algorithm.
Vol. 55, No. 5, May 2009, pp. iv–vi
issn 0025-1909 eissn 1526-5501 09 5505 00iv
doi 10.1287/mnsc.1090.1039
© 2009 INFORMS
Management Insights
Blockbuster Culture’s Next Rise or Fall: The Impact
of Recommender Systems on Sales Diversity (p. 697)
Daniel Fleder, Kartik Hosanagar
The last ten years have seen an extraordinary increase
in the number of products available. This trend is
part of the “long tail” phenomenon, and many believe
that it could amount to a cultural shift from hit
to niche goods. A difficulty that arises, however, is
how consumers will find their ideal, niche products
among myriad choices. Recommender systems are
one solution. These systems use data on purchases
and user profiles to identify which products are best
suited to each user. Although recommenders have
been assumed to diversify choice, we show why some
systems may do the opposite. Recommenders can
create self-reinforcing cycles in which popular items
are recommended more, recommended items are pur-
chased more, purchased items are recommended even
more, and so on. These cycles reduce diversity. Con-
sequently, consumers and niche producers may be
underserved if there exist better product matches out-
side of the hits, and retailers may find that they offer
the right assortment but their recommender system
may be promoting a narrow range of products. We
recommend that managers consider design modifica-
tions to ensure that their recommender system limits
these popularity effects and promotes exploration.
On the Value of Commitment and Availability
Guarantees When Selling to Strategic
Consumers (p. 713)
Xuanming Su, Fuqiang Zhang
Product availability plays an important role in attract-
ing consumer demand. Despite technological and
managerial advances, industry evidence shows that
stockouts are a common phenomenon and prod-
uct availability remains a key issue in marketing
and operations. In environments in which consumers
make choices based on product availability, we pro-
pose two strategies that firms can use to improve
profits. First, firms can make upfront commitments
to consumers that at least a certain quantity will
be stocked. Second, firms can provide availability
guarantees to compensate consumers in the event of
stockouts. Interestingly, firms may have an incentive
to overcompensate consumers during stockouts. To
attain maximum possible profits, we show that firms
need to use both strategies in conjunction.
An Optimal Contact Model for Maximizing Online
Panel Response Rates (p. 727)
Scott A. Neslin, Thomas P. Novak, Kenneth R. Baker,
Donna L. Hoffman
This paper develops and field tests a model for maxi-
mizing response rates for online survey research pan-
els. The model includes several important features:
(i) it accounts for the “state” of each panelist (e.g., pre-
vious response rate); (ii) it can plan for several studies
at a time; (iii) it recognizes that current decisions may
influence future response rates; (iv) it allows the user
to stipulate the desired sample size and demographic
makeup for the sample; and (v) it anticipates growth
in the panel. In a field test conducted for four stud-
ies, the model yields an average response rate of 43%
per study, compared to 25% for the heuristic currently
used by an online panel’s manager and 14% for ran-
dom selection. These results suggest that managers
could use the model to improve upon current prac-
tice either by obtaining the same sample size while
soliciting fewer panelists (thus avoiding panelist
“burnout”) or by increasing the sample size while
soliciting the same number of panelists (thus provid-
ing smaller standard errors and hence more accurate
Contagion of Wishful Thinking in Markets (p. 738)
Nicholas Seybert, Robert Bloomfield
How does one person’s behavior affect the decisions
of others when all parties are making risky decisions
to increase their wealth? We show that investors in
a stock market have a tendency to engage in “wish-
ful betting,” where they invest or bet as if desirable
outcomes are unreasonably likely. When one investor
engages in wishful betting and purchases additional
shares of stock (as if he believes shares are under-
valued), other investors may fail to adjust for this
bias, thereby leading them to hold unreasonably opti-
mistic beliefs, which we term “wishful thinking.”
Wishful thinking could occur in a variety of con-
texts, including managerial decisions influenced by
competitor behavior, and individual career path deci-
sions influenced by peer behavior. The results of our
studies suggest that people should be cautious when
interpreting others’ behavior; otherwise, they may
unwittingly sacrifice their own wealth when making
Management Insights
Management Science 55(5), pp. iv–vi, © 2009 INFORMS v
Quasi-Robust Multiagent Contracts (p. 752)
Anil Arya, Joel Demski, Jonathan Glover, Pierre Liang
Incentive contracting has been a much-studied topic
in information economics. However, it has been noted
that the theoretically derived optimal contracts are
often at odds with observed practice in that they
are highly fine-tuned to the details of the envi-
ronment. Many of these theoretically optimal con-
tracts would perform quite poorly if the environment
were even slightly different from that assumed. This
paper attempts to contribute to the emerging the-
ory of robust contracts in the hope that such con-
tracts will help us better understand observed prac-
tice. A distinguishing feature of our approach is
that we assume a manager must choose the design
of a contract mechanism before some key informa-
tion is known about the environment. We use our
method to provide insights into an auction that has
to be designed for a variety of bidders and bidder
Multiple Sourcing and Procurement Process
Selection with Bidding Events (p. 763)
Tunay I. Tunca, Qiong Wu
We study the process selection problem of an indus-
trial buyer who employs online reverse auctions for
procurement. We compare two types of procurement
processes: (1) simple reverse auctions (“single-stage”
processes) and (2) processes where the buyer makes
additional price-quantity adjustments with the win-
ning suppliers after the auction (“two-stage” pro-
cesses). If there is a large number of bidding sup-
pliers and production is not scalable (i.e., capacity is
rigid), then we find that single-stage procurement is
preferred. However, the two-stage process tends to be
relatively more attractive as the number of bidders
decreases or as capacity becomes more scalable (i.e.,
an increase in quantity does not generate a consider-
able increase in the per-unit cost).
Information Sharing and Order Variability Control
Under a Generalized Demand Model (p. 781)
Li Chen, Hau L. Lee
Information sharing between partners is one means
to improve supply chain performance, e.g., through
mitigation of the bullwhip effect. However, when
a retailer shares point-of-sales data, the supplier
can fully exploit this information only if the sup-
plier also has knowledge of the characteristics of
the demand process and the retailer’s order policy.
How can a supplier realize the value of informa-
tion sharing when such knowledge is lacking? Our
paper shows that this can be achieved by having the
retailer share its projections of future orders and their
A Generalized Approach to Portfolio Optimization:
Improving Performance by Constraining Portfolio
Norms (p. 798)
Victor DeMiguel, Lorenzo Garlappi,
Francisco J. Nogales, Raman Uppal
We provide a general framework for finding portfo-
lios that perform well out-of-sample in the presence
of estimation error. This framework relies on solv-
ing the traditional minimum-variance problem but
subject to the additional constraint that the norm
of the portfolio-weight vector be smaller than a
given threshold. We show that several established
approaches in the literature are actually special cases
of our framework. We use five data sets to compare
the out-of-sample performance of our method with
10 known strategies and find that our method often
yields a higher Sharpe ratio.
Optimal Policies and Approximations for a Bayesian
Linear Regression Inventory Model (p. 813)
Katy S. Azoury, Julia Miyaoka
We consider the inventory management problem
when demand is estimated using a regression model.
We assume the regression parameters are unknown,
and a Bayesian approach is used to update the dis-
tribution on the regression parameters as new infor-
mation becomes available. Within our framework we
identify the optimal inventory policy. However, it is
computationally complex to implement, so we pro-
pose heuristic policies and demonstrate that they
yield near-optimal performance.
Information Market-Based Decision Fusion (p. 827)
Johan Perols, Kaushal Chari, Manish Agrawal
In many decision-making scenarios, such as fraud
detection and bankruptcy prediction, the decisions of
multiple human experts and/or software are fused to
determine the overall decision. This paper provides
an innovative approach for decision fusion based on
information markets. Our computational results indi-
cated that our approach is superior to other existing
approaches. This information market-based method
can help organizations lower costs and facilitate new
decision-making systems that combine the expertise
of humans and software.
Private Network EDI vs. Internet Electronic
Markets: A Direct Comparison of Fulfillment
Performance (p. 843)
Yuliang Yao, Martin Dresner, Jonathan Palmer
How can purchasing organizations decrease cycle
times and improve order fulfillment? The key to
these performance improvements may lie in the sup-
ply chain technology used for transaction exchanges.
Using a data set comprised of 2.8 million transactions
Management Insights
vi Management Science 55(5), pp. iv–vi, © 2009 INFORMS
placed through the U.S. government’s Federal Sup-
ply Services, we provide a direct comparison between
private network electronic data interchange (EDI) sys-
tems and Internet-based electronic market systems.
We find that when purchasers use the Internet-based
electronic market, cycle times are reduced by two
days and complete orders fulfilled are increased by
two percentage points, compared to the competing
EDI system.
Loss Functions in Option Valuation: A Framework
for Selection (p. 853)
Dennis Bams, Thorsten Lehnert, Christian C. P. Wolff
We investigate the importance of different loss func-
tions when estimating and evaluating option pricing
models. Our analysis shows that it is important to take
into account parameter uncertainty because this leads
to uncertainty in the predicted option price. We find
strong evidence to support the idea that the absolute
pricing error criterion may serve as a general-purpose
loss function in option valuation applications. At the
same time, we provide a first yardstick to evaluate the
adequacy of the loss function. This is accomplished
through a data-driven method to deliver not just a
point estimate of the pricing error, but a distribution.
Additive Utility in Prospect Theory (p. 863)
Han Bleichrodt, Ulrich Schmidt, Horst Zank
Decision making in managerial environments where
alternatives consist of risky multiple-attribute out-
comes is becoming a very difficult task because of
the complex way in which individuals evaluate risks.
Utility measurement tools have been developed for
single-attribute outcomes but those tools may be
inappropriate for multiple-attribute outcomes if loss
aversion is attribute specific. For example, when indi-
viduals evaluate job offers, the effect of loss aversion
relating to a drop in salary income may be of different
magnitude than the effect of loss aversion relating to
how onerous the job is. We show how utility measure-
ment can be improved if loss aversion is understood
as an attribute-specific feature, and we provide new
decision models that improve the empirical toolbox
of decision makers and managers.
... They use 974 VARX models similar to Eq. (2.14). An interesting perspective is provided by Pauwels (2007). He finds that com- promotions or discounts because the product will be in high demand regardless. ...
... Natter et al.(2007) present a decision support system for dynamic retail pricing 1369 and promotion planning. Their weekly demand model incorporates price, reference 1370 price effects, seasonality, article availability information, features and discounts.1371They ...
... advantage of this approach is that decisions are made on a periodic basis using the 1440 exact data from the customer's recent purchase history, not their estimated data (see 1441 alsoNeslin et al. 2009). However, the optimization has to be re-run each decision 1442 period. ...
Full-text available
Firms spend a significant part of their marketing budgets on sales promotions. Since the impact of promotions on sales is usually immediate and strong, promotions are attractive to results-oriented managers seeking to increase sales in the short term. This chapter discusses models for measuring sales promotion effects. Part I focuses on descriptive models, i.e. models that describe, analyze, and explain sales promotion phenomena. We start by discussing models for analyzing the immediate impact of promotions and decomposing the resulting sales promotion bump into a variety of sources. Next we examine what happens after the immediate bump, and describes models for measuring feedback effects, reference price effects, learning effects, permanent effects, and competitive reactions. Next we turn to descriptive models for promotions aimed at retailers (“trade promotions”), and discuss forward buying and pass-through. In Part II we discuss normative models, i.e. models that tell the decision maker what is the best (profit maximizing) decision on sales promotion activities. We cover models for planning promotions to consumers as well as decision models on trade promotions for manufacturers, and normative retailer models for optimal forward buying and pass-through. Part III concludes this chapter with a summary, practical model guidelines, and directions for future research.
... Owning one's own research panel provides researchers with much greater control over the signup and research process, but the trade-off is in the time and financial cost required to build and maintain the panel. Maintaining a research panel, however, does allow researchers to investigate problems involving the operation of the panel itself, such as methodology for devising optimal contact strategies that maximize survey cooperation rates of panel members (Neslin, Novak, Baker, and Hoffman 2009). The basic operation of the eLab City Panel is described in the next section. ...
... In two cases, each of the following predictors is significant: women are more likely to complete (columns c and d), older avatars are more likely to complete (columns a and e), and those panelists with a higher education are more likely to complete (columns e and f). The weak prediction of study completion by demographic characteristics is not surprising, as completion of previous studies and time since last invitation have been found to be stronger predictors of completion rates than user demographics (Neslin, Novak, Hoffman and Baker, 2009). ...
The eLab City project in Second Life is a laboratory environment for the study of user behavior in virtual worlds. This paper describes the origin and development of the eLab City project, which includes virtual infrastructure constructed in Second Life, a panel of Second Life users who have agreed to participate in research studies, tools for observational data collection, and procedures for fielding research projects. The eLab City panel is described in detail, with discussion of the panel signup process, recruitment, and panelist demographics. A series of research studies have been fielded with eLab City panelists, and cooperation rates for these studies are presented, together with comparison to Web-based studies. We conclude with a discussion of lessons learned and next steps for this research project.
... Similar to Study 1, Study 2's sample consisted of a crosssection of small businesses. Two hundred sixty small businesses were recruited using Qualtrics panel services, which have become an established research sample pool method (Jiménez & Mendoza, 2013;Neslin et al., 2009). Ninety-four percent of companies had fewer than 50 employees. ...
Full-text available
Social media and the Web have empowered consumers to share a myriad of information online about companies with reduced physical and psychological costs, rendering businesses responsible for acknowledging customer demands in order to obtain approval. In the face of such perceived scrutiny, this research investigates in two studies how proactive webcare can foster positive online interactions with customers as well as improve a firm’s presence on the web. Using a mixed method approach to explore these phenomena, findings indicate that small businesses believe that socially prescribed perfectionism has manifested in the online environment; however, the effective use of webcare is positively associated with the promotion of consumer-generated content sharing which positively influences engagement and online reputation management.
... Out of the 62 responses, only 39 were valid (there were 23 invalid responses 1 ), representing a response rate of 25%. This is consistent with reported response rates for online panels [66]. The analysis of the sample of survey respondents identified several distinct expert groups by their primary link to the biofuel sector. ...
Social dimension appears to be the least developed of all dimensions of sustainability, not receiving the same attention as environmental or economic dimension. While biomass utilization is considered to have considerable impact on the social well-being of farmers and local communities, a better understanding of its social sustainability is urgently needed. The process for determining social issues, however, is subject to relatively arbitrary decisions, and lacks comprehensive structure. Social issues must be based on those social objectives and indicators that can be empirically measured and analyzed using at the existing level of knowledge and data available. This study, therefore, aims to identify the most important and relevant social and governance issues for the biofuel sector, and also to determine the issues for which reliable data and practical methods may become available and ultimately simplified for understanding by stakeholders. The sugarcane biojet fuel supply chain in Brazil was used as a case study with a research design of two steps: literature review and expert survey. From the literature review, 13 social issues and 5 governance issues were selected for inclusion in the expert survey. The survey results showed that highly relevant issues were generally perceived as highly important. Furthermore, very practical issues were also perceived as very reliable and simple issues. It was concluded that future research should mostly focus on quantitative assessment of human health and safety, labor rights, working conditions, which were perceived very important but less reliable, practical, and simple. Moreover, this study showed that all governance issues are certainly regarded as important for sustainability, but insufficiently recognized in conventional sustainability assessment schemes. The current certification schemes cover only a limited number of social issues and require addressing social issues more broadly. Learning from this study helps decision makers to extend understandings of the social dimension of sustainability.
... A total of 381 responses (including invalid responses) were received, representing a response rate of 8%. This is consistent with the reported response rates for online panels ( Neslin et al. 2009). This sample size is statistically significant with 95% confidence level and 10 confidence interval. ...
Conference Paper
Full-text available
Improving building energy efficiency is one of the best strategies to reduce building energy consumption. Recent studies emphasized the importance of occupant behavior as key means of enhancing building energy efficiency. It is critical that while we strive to improve the energy efficiency of buildings through the understanding of energy use behavior that we also understand the values (such as thermal comfort, indoor air quality, productivity) of building occupants, how these values may impact energy use behavior, and how we can improve energy efficiency without negatively impacting these values (i.e., while maintaining the satisfaction levels with these values). This paper focuses on presenting the authors work in (1) identifying potential occupant values that may impact energy use behavior and energy consumption in residential buildings, (2) discovering actual building occupant values and the importance levels of these values to residential building occupants, and (3) discovering the current satisfaction levels of residential building occupants with these values. The discovery of actual occupant values and current satisfaction levels was conducted using an online survey. A randomly selected set of 310 residential building occupants in Arizona (AZ), Illinois (IL), and Pennsylvania (PA) were surveyed using an online questionnaire. The paper discusses the value discovery, questionnaire design, survey results, results analysis, and conclusions. The results showed similarities and differences across occupants in AZ, IL, and PA in terms of what they value in buildings as well as their current satisfaction levels with these values.
... However, we do not observe the actual response rate of retention offers by customers, which would allow us to estimate the response probability of a given action, e.g. by using repeated observations of customers' response to past promotions (Neslin et al. 2009). Alternatively, a preliminary field experiment with a control (no retention offer) and treatment group (targeted at random with a retention offer) would allow the estimation of heterogenous treatment effect model (Provost and Fawcett 2013, p. 287), e.g. using uplift random forests (Ascarza 2016). ...
Full-text available
Customer defection threatens many industries, prompting companies to deploy targeted, proactive customer retention programs and offers. A conventional approach has been to target customers either based on their predicted churn probability, or their responsiveness to a retention offer. However, both approaches ignore that some customers contribute more to the profitability of retention campaigns than others. This study addresses this problem by defining a profit-based loss function to predict, for each customer, the financial impact of a retention intervention. This profit-based loss function aligns the objective of the estimation algorithm with the managerial goal of maximizing the campaign profit. It ensures (1) that customers are ranked based on the incremental impact of the intervention on churn and post-campaign cash flows, after accounting for the cost of the intervention and (2) that the model minimizes the cost of prediction errors by penalizing customers based on their expected profit lift. Finally, it provides a method to optimize the size of the retention campaign. Two field experiments affirm that our approach leads to significantly more profitable campaigns than competing models.
This article provides a novel method to optimize the profits of proactive retention campaigns using a machine learning approach to causal inference.
Customer relationship management (CRM) is becoming a very hot topic nowadays in academia and business environments. Indeed, companies are constantly searching for new innovative ways to create or maintain their competitive advantage. Due to the recent advances in Internet and technology, CRM predictive analytics is becoming an important tool in the toolset of the marketer. It is the practice of using the huge volumes of historical customer data to predict future customer behavior. This chapter introduces the reader to the shift towards a data-driven customer centricity approach, where marketers act upon what they know, rather than upon what they think.
Full-text available
This paper was chosen for an Emerald Literati award. For the next six months, it may be downloaded FREE at: . In this paper, the focus is on the evolution of online survey research from 2005 to the present, as well as a look at what is ahead for survey research. Online surveys not only continue to grow in popularity and to mature as a research technique, but new technologies have also changed survey methodologies—and possibilities. While earlier research and interest as to online surveys were driven by the potential benefits of doing research online, the growth of online surveys has further brought to the forefront some limitations and pitfalls of doing surveys online.
In the last 20 years the marketing literature has seen a sharp increase in the number of papers reporting findings from field experiments. This can be partly explained by the ease of conducting field experiments in Internet settings. However, we have also seen an increase in field experiments in physical stores and other non-Internet settings. While many of these papers focus on pricing and advertising topics, there are also a broad range of other topics represented, including several papers that use field experiments to provide model-free validation of optimization models. We review the requirements to publish a field experiment paper in the marketing literature. We also identify topics that remain relatively understudied. In particular, there is a notable absence of papers studying channel relationships or business-to-business markets. Perhaps more surprisingly, there is also a lack of papers investigating the feasibility of using field experiments to optimize marketing decisions.
Full-text available
Customized communications have the potential to reduce information overload and aid customer decisions, and the highly relevant products that result from customization can form the cornerstone of enduring customer relationships. In spite of such potential benefits, few models exist in the marketing literature to exploit the Internet's unique ability to design communications or marketing programs at the individual level. The authors develop a statistical and optimization approach for customization of information on the Internet. The authors use clickstream data from users at one of the top ten most trafficked Web sites to estimate the model and optimize the design and content of such communications for each user. The authors apply the model to the context of permission-based e-mail marketing, in which the objective is to customize the design and content of the e-mail to increase Web site traffic. The analysis suggests that the content-targeting approach can potentially increase the expected number of click-throughs by 62%.
Full-text available
Rhenania, a German direct mail-order company, turned its catalog mailing practices around within one year and consequently moved up in market position from number 5 to number 2. A dynamic multilevel modeling (DMLM) approach uses elasticities to determine the optimal frequency of catalog mailings, a customer-segmentation approach allows for optimization of mailings, and a recency, frequency, monetary-value (RFM) segmentation in combination with a chi-square automatic interaction detection (CHAID) algorithm determines when customers should receive a reactivation package—as opposed to a catalog—to optimize mailing efficiency further. The DMLM approach was so effective that Rhenania acquired two competitors (one a subdivision of Springer Verlag).
Full-text available
The authors express their gratitude to Sanyin Siang (Managing Director, Teradata Center for Customer Relationship Management at the Fuqua School of Business, Duke University); research assistants Sarwat Husain, Michael Kurima, and Emilio del Rio; and an anonymous wireless telephone carrier that provided the data for this study. The authors also thank participants in the Tuck School of Business, Dart-mouth College, Marketing Workshop, for comments and the two anony-mous JMR reviewers for their constructive suggestions. Finally, the authors express their appreciation to former editor Dick Wittink (posthumously) for his invaluable insights and guidance. This article provides a descriptive analysis of how methodological factors contribute to the accuracy of customer churn predictive models. The study is based on a tournament in which both academics and practitioners downloaded data from a publicly available Web site, estimated a model, and made predictions on two validation databases. The results suggest several important findings. First, methods do matter. The differences observed in predictive accuracy across submissions could change the profitability of a churn management campaign by hundreds of thousands of dollars. Second, models have staying power. They suffer very little decrease in performance if they are used to predict churn for a database compiled three months after the calibration data. Third, researchers use a variety of modeling "approaches," characterized by variables such as estimation technique, variable selection procedure, number of variables included, and time allocated to steps in the model-building process. The authors find important differences in performance among these approaches and discuss implications for both researchers and practitioners.
Full-text available
Catalog sales are among the fastest growing businesses in the U.S. The most important asset a company in this industry has is its list of customers, called the house list. Building a house list is expensive, since the response rate of names from rental lists is low. Cash management therefore plays a central role in this capital intensive business. This paper studies optimal mailing policies in the catalog sales industry when there is limited access to capital. We consider a stochastic environment given by the random responses of customers and a dynamic evolution of the house list. Given the size of real problems, it is impossible to compute the optimal solutions. We therefore develop a heuristic based on the optimal solutions of simplified versions of the problem. The performance of this heuristic is evaluated by comparing its outcome with an upper bound derived for the original problem. Computational experiments show that it behaves satisfactorily. The methodology presented permits the evaluation of potential catalog ventures thus proving useful to entrepreneurs in this industry.
Full-text available
When a demand pattern is dominated by a seasonal effect, the concept of a steady state solution can be used in two ways in aggregate production planning. First, general policy recommendations can be made concerning the use of seasonal workforce changes versus overtime and seasonal inventory. Second, the results can be used to provide ending conditions in an intermediate range planning algorithm with a moving horizon. These ideas are explored using a linear programming model, making use of known planning horizon properties. A simulation experiment tests the efficacy of short horizon deterministic models in a stochastic environment, and demonstrates that ending conditions, derived from the steady state model, improve decisions under a variety of conditions on costs and horizon length.
An experimental study was designed to investigate the efficiency of decisions obtained from optimizing a finite, multiperiod model and implementing (structure is parallel with “optimizing” above) those decisions on a rolling basis. The results of the study suggest that rolling schedules are quite efficient and also that they point to some important design issues in model-building for production planning.
We present a classified bibliography of the literature in the area of forecast, solution, and rolling horizons primarily in operations management problems. Each one of over 200 selected papers is categorized on five dimensions that identify the horizon type, the model type (deterministic or stochastic), the sources of the horizon, the methods used to obtain horizon results, and the subject area of the paper. The majority of the papers treat dynamic problems in inventory management, production planning, capacity expansion, machine replacement, and warehousing. We discuss the relationship of the horizon results with the theory and practice of rolling-horizon procedures and future research directions.