TITLE Price Behavior ( and Forecasting and Elasticities ) in the Pulp and Paper Industry

The goal of the proposed project is to identify economic and regulatory factors that have generated price movements in the pulp & paper industry in order to meet a tactical industry objective of developing improved models for understanding pricing behavior and a strategic industry objective of ensuring the long term economic health of the industry. The proposed project will employ advanced econometric techniques in order to develop improved paper and pulp pricing models. In particular, regression-based analysis is employed to statistically model those factors that are found to influence prices and to estimate the sensitivity of prices to changes in these determinants. Complementing this, historical time series data will be used to develop pricing behavior models in order to identify the stochastic process that governs price movements. Finally, the project will combine regression analysis with time-series methods to develop more efficient pricing behavior models. In addition to price behavior analysis, this project will also explore the feasibility of implementing a real-time market demand forecast for individual producers. By integrating product demand forecasting tool into production and management processes, producers will be able to adjust production levels in anticipation of market fluctuations, thus avoiding involuntary inventory buildups. This would enable firms to support product prices at a more efficient level. The proposed project will contribute to research by deepening our understanding of market interactions and price behaviors in the pulp and paper industry. The project will also further the strategic objectives of the paper and pulp industry by providing important insights on those economic determinants that generate price movements and by providing improved pricing models intended to explain price fluctuations with an implication on future price movement. The desired outcomes of the project include a series of research papers and reports on industry structure, pricing behavior, as well as real-time market demand forecasting tools for individual producers. The proposed project will last for three years. The first year will focus on an industry survey, a case study, an evaluation of existing work on price analyses, and the construction of a preliminary model for the containerboard sector of the industry. The research team includes faculty members from Georgia Tech and the IPST, as well as graduate and undergraduate students. With a broad and in-depth knowledge of econometric and statistical methods, industrial organization, systems engineering, and paper manufacturing technology, the interdisciplinary team possesses the abilities that are necessary for meeting the important research and industry objectives of the proposed project. Our first year budget will be $90,000. The second year has a minimum $80,000 budget and the third year minimum budget is $50,000. PURPOSE, GOALS & LITERATURE REVIEW Significant and unpredictable paper and pulp price movements have had a number of serious consequences for the pulp & paper industry, including excess capacity, unintended inventory build-up, and financial losses. And in the long term, unanticipated price behavior will threaten the economic viability of the industry. The primary focus of the proposed project is the pricing behavior of pulp and paper products. Our purpose is to explain price behavior, identify the causative factors, estimate various elasticities, and build statistical pricing behavior models for various segments of the industry. The goal is to advance academic research on pricing using modern econometric methods, and to enhance the industry’s understanding of past and, by extension, future price movements. This project will explore the feasibility of establishing real-time market forecasting models that would enable individual producers to avoid involuntary inventory buildup, which would reduce the pressure on firms to cut prices during a market downturn. I. Current Forecasting Practices and Existing Studies Understanding pricing behavior is among the most important issues for the pulp and paper industry. Information on prices are essential for budgeting, project financial assessment, contract negotiations and capacity planning. However, explaining price movements poses a formidable challenge. Fundamentally, market prices are determined in a system of equations that include demand and supply side variables interacting within a specific market structure. Because many variables in this system are endogenous, it is by no means an easy task to accurately model price movements and, therefore, to evaluate where prices may be headed. In the pulp and paper industry, Resource Information and Systems Inc. (RISI), the Jaakko Pöyry Group, and the Pulp and Paper Forecaster by Miller Freeman Inc. generate a detailed set of forecasts. They publish regular forecasts of demand and prices for almost all pulp and paper products in different regions. While such information has been helpful to the industry, industry experts concur that these forecasts have not been satisfactory. More specifically, the underlying pricing models neither focus on causative factors nor do they provide information on price sensitivities due to changes in the causative factors. In addition, econometric techniques and methodologies have significantly advanced in recent years, yet many of the existing models of price behavior fail to reflect these advances. Academic research on pricing in the pulp and paper industry has almost exclusively focused upon the relationship between exchange rates and prices. For example, Alavalapati et al. (1997) uses co-integration analysis to investigate the effects of the Canada-U.S. exchange rate and U.S. pulp price on the price of Canadian pulp. Hanninen and Topinen (1999) estimate the pass-through effects of exchange rate variations on Finnish pulp and paper exports. Uusivuori and Buongiorno (1990) investigate the short-run and long-run effects of changes in exchange rates on U.S. imports of paper from Finland and Sweden. These studies certainly further the understanding of how exchange rates influence price movements, but they do not develop a general framework that focuses on determinants of price behavior and their associated elasticities. Nebebe and Kira (1992) use Bayesian techniques and least squares approaches (e.g. seemingly unrelated regression methods) to examine the long-term elasticities of demand for pulp and paper products in Canada in selected regions. Singh and Nautiyal (1984) study the factors that affect pulp and paper prices. These studies provide useful estimates of price and income demand elasticities and identify those factors that affect price. For our purposes, Singh and Nautiyal is the most relevant but its focus is on the Canadian pulp and paper industry and on the Canadian market structure in general, and incorporates assumptions that may not carry over to the U.S. pulp and paper industry. In sum, there are very few studies on prices for the pulp and paper industry in the U.S. The proposed project attempts to fill this gap by developing alternative models of pricing behavior with the aim of understanding the forces that have shaped industry prices in the past and, therefore, will provide insights on future prices for specific pulp and paper products. II. Understanding Aggregate Price Movements Regression-based econometric analysis and the Box-Jenkins statistical methods (Box and Jenkins, 1970) will be employed for analyzing prices behavior in the pulp and paper industry. Econometric models use explanatory variables to analyze price changes and are very useful for identifying the important factors that affect prices and for estimating various elasticities. However, these models are less useful if one is concerned about quantitative estimates of future prices since this requires the analyst to either forecast or assume values for all relevant explanatory variables, complicates the forecast since the uncertainties associated with forecasting explanatory variables are passed onto the price forecasts. In contrast, Box-Jenkins methods identify the data generation process that governs price movements. The Box-Jenkins approach develops Autoregressive Integrated Moving Average (ARIMA) models that do not typically rely on other explanatory variables for characterizing price behavior. ARIMA models uses information on past prices to identify the data generation process that governs price movements. Such an approach works in two ways. First, intervention analysis focuses upon whether an external event, such as the onset of an energy crisis or passage of more restrictive environmental laws affecting the industry, alters the data generating process. Second, ARIMAX models, which incorporate independent variables, can be estimated, in order to explicitly control for the effects of a limited number of explanatory variables. ARIMA models are useful for identifying the stochastic process and, accordingly, for developing quantitative estimates of future prices without any knowledge of explanatory variables or other technical problems associated with regression models. Although ARIMA models require much less data collection, the approach cannot help explain why prices move in a particular way and what the causative factors are. Nor can it provide information on various elasticities of price responses. Still, intervention analysis and ARIMAX models are useful in exploiting the small data requirements and yet provide some insights on those factors important to price behavior. Because each approach has its advantages and disadvantages, this project will use regression based and Box-Jenkins based methodologies to study price behavior. Ultimately, these two approaches will be combine to produce a much better understanding of pricing behavior in the industry than would be possible with either technique alone. More specifically, when using regression analysis, the price variation unexplained by the explanatory variables will be left to the error term. Given the time-series nature, these errors are likely to be correlated. Such correlation offers useful information for the purpose of forecasting because current errors inform us about future errors. By constructing an ARIMA structure for the regression errors, we can obtain more efficient forecasts. In constructing our regression models for prices, we will include usual demand and supply side variables, such as economic activity and cost variables. More importantly, we will investigate the effect of inventories on prices, which has not been studied in previous work. Clearly, price movements are affected by demand, production, and inventories. Price changes generally start from involuntary inventory building-up or running-down. Therefore, inventory effects have important implications for price movements. Furthermore, we will also control in the price model for the effect of productivity increases caused by technological progress, including the effect of the Internet. It can be anticipated that a number of technical problems will arise in estimating a price model. In particular, the stationarity of prices should be first examined and tested before running any regressions. Also, a number of variables might be endogenous and thus instrumental variable estimation may need to be applied. Other problems, such as serial correlation and heteroskedasticity, will relatively easy to statistically test and, if necessary, correct. Finally, in an effort to evaluate whether our models have adequately captured pricing behavior in the industry, point and interval forecasts, and their associated confidence intervals, will be generated. A number of criteria will be used to evaluate the models and the forecasts, including ex post forecast and comparison with existing industry forecasts. III. Firm-Level Market Forecasting Tools Complementing our analysis of pricing behavior at the industry level, we will also develop market demand forecasting tools at the level of firms to guide an individual firm’s production during a market swings (upturns and downturns). In a downturn, for example, this will help the firm to support its prices and improve efficiency. More specifically, over-capacity and involuntary inventory buildup have been the main pressure for price drops when facing market fluctuations. If each producer can anticipate the market demand and respond prior to the downturn, over-production and excessive inventory can be largely avoided. Therefore, it would be an effective means to help support price levels if all firms adopt modern market demand forecasting tools and use them in real-time production planning and inventory management. This proposed project will look into the feasibility of providing such real-time, automatic (and even web-driven) market demand forecasting tools tailored to specific firms and products. In the last several years, two strategies have been proven to be the most effective in supporting price levels, the industry’s restraint on capacity expansion, and the fact that producers are more aggressive in taking down time in response to weakness in demand. Downtime means running machines more slowly, shutting machines temporarily, and shutting machines indefinitely. Based on the “Pulp and Paper Forecaster” (2000), the North American containerboard producers have been leading the strategy of taking downtime and others are following their path. If each firm adopts real-time market demand forecasting in planning its production prior to market changes, the firm can avoid or largely reduce involuntary inventory buildup. Therefore, prices will move in a more natural way. This proposed project will explore such a possibility by building a demonstration model for selected companies. This kind of firm-level demand forecasting, in general, does not require great accuracy and should be relatively easy to build, e.g., using exponential smoothing. The advantage for such low-cost firm-level market demand models is that they are easy to use and can be web-driven. In each period, market demand forecasts will be updated when new information is available. Moreover, the process can be made to run automatically with an integrated mechanism to track forecasting performance. If systematic errors occur, the process will generate warning signals for human intervention to adjust the forecasting routines.