Forest2Market has demonstrated an unprecedented 95 percent accuracy rate with forecasting.

Forest2Market's stumpage Price 4Cast relies upon a series of econometric models (see Figure A). An econometric model is a mathematical representation of relationships in an economy – in this case a local stumpage market economy – expressed as equations. The equations explain how one economic variable could change as a result of changes in other key factors. Econometric models are developed using statistical techniques to identify key factors and describe their relationships to other variables. The models are built looking at past data and relationships between selected economic factors. To describe models and their likely changes, we group factors into one of four effects: a general economy effect, a local market effect, an industry sector effect, and a seasonal and momentum effect (see Table A). We strive to have an individual model “balanced” among these four classes of effects.

Figure A. 4Cast Models Schematic

Econometric models are frequently criticized for using the past to predict the future. But we have observed that experience is one of the more valuable elements in decision making. In essence, these types of statistical models are a way to summarize “experience” and apply that experience to future situations. As is the case with any model, econometric models are not perfect, but they can deliver useful insights when forecasting changes in the economy.

Table A. Model Factors Listed by Major Effect

When describing probable forecasted changes, it is convenient to speak of one factor “causing” another to react in a certain way. The reality is most econometric models are correlative in nature, not causative; the 4Cast is no exception. A correlative model means two factors show a certain tendency to move in a definable pattern with respect to each other. This relationship could be because one factor directly bears on the other, or it could be both factors are responding to yet another primary “cause” in the system.

In developing 4Cast models, our emphasis is on forecasting. In contrast, academic economists who use techniques similar to the ones we employ often focus on isolating the primary factors that are correlated with, and may account for, past changes in the variable being described by the model. The percentage of historical variability (both above and below the historical average) explained is a commonly used measurement of how well a statistical model performs (see Figure B).

While a few key factors may explain a large share of the historical variability, it is unusual for that explanatory share to rise above 70%. That result may be sufficient if the purpose is to understand how economies work, but it is not very good if one is trying to forecast future direction and price levels for supply chain applications. Our modeling efforts are focused on including factors that collectively explain a high degree of the historical variation seen in wood fiber product pricing in the local market, behave within the model in a logical fashion, are consistent with single variable correlations to the dependent variable, and offer reasonable opportunities for making accurate forecasts (see Figures C1 and C2). Many competing forecasting firms provide no indication of how well their models fit the historical data; by contrast, one can see how well our models fare in that regard in every 4cast we publish.

Economic systems operate as many systems do, maintaining balance and having feedback cycles to correct imbalances when they occur. We integrate these elements in our models in a variety of ways:

• We create variables that express interactions among the foundational factors. For example, trying to explain price behavior via an exchange rate, GDP and housing starts may yield results that are marginally enlightening. But a second-tier factor that incorporates an interaction among those three “base” factors often does a much better job of explaining (and predicting) price behavior;
• We design our model system to reflect that sub-regional markets often trade off relative to one another due to shifting demand pressures;
• Relationships between factors can change depending on their timing in relation to each other. This latter point can be great news when forecasting. For example, if it turns out the price in any given month is correlated with a factor that occurred four months earlier, that means something that occurred three months ago will indicate where the price is headed next month; what is going on right now is providing insight into prices four months from now.

The complexity of economic systems can mean that models alone sometimes fail to accurately portray feedback mechanisms and responses. Experience is important in human decision-making, but it alone doesn’t guarantee a good final decision. Likewise, we test our models’ results against a series of logic checks and modify outcomes when it seems a past relationship is either too robust or inadequate in light of other influences and pressures.

Different markets have different characteristics. Examples of a few distinguishing elements between markets include the types of manufacturing facilities, land ownership patterns, transportation infrastructure, and ground conditions. These differing characteristics give rise to market-specific models even though the same collection of factors is considered for every market’s models.

The forecast is a unique industry offering with a near-term supply chain application focus – average monthly prices for the next 24 months. The stumpage prices being forecasted are based on F2M’s market transaction data, not market opinions. This means there is a direct benchmark value in the forecast rather than using it as a relative index of market dynamics. The models are local market focused, which means 4Cast results relate to actual prices being paid for wood fiber in a real market – not an average for some hypothetical geographic area that does not function as a market. Along with relating the impact of global and national effects on local markets, the models consider real world factors, including weather and ground conditions that influence day-to-day operations and decisions. Finally, the models are updated quarterly, reflecting market dynamism that is a reality in today’s globally linked economy.