Traditional Forecasting and Modeling Methods
Trending, Extrapolation and Curve Fitting Methods – are typically used when the forecast time frame is short to medium term and there is sufficient evidence that forecast inflection points do not exist in the time frame. The most common is the exponential (growth) curve and may be applied not just to the primary time series but to second or third differences such as the trend in rate of change.
Adoption & Penetration Models – such as Fisher-Pry, Gompertz or the Bass Diffusion Model are the method of choice when there is sufficient evidence that the historic and forecast time frame of the model will include inflection points. The scope of these models is usually total market, and the objectives are to predict market penetration, location of the inflection points, and take-up times.
Causal and Multivariate Methods – are utilized when there are multiple known causal influences driving or constraining the market. The two most common forms of this approach are:
- Total Market models that are functions of a combination of economic, demographic, sociological and technological factors. These methods commonly require primary and/or secondary research, plus statistical analysis, to identify the causal variables and how they are related to the forecast variables.
- Market Share methods that allocate based on the identification and importance of competitive advantage factors. and estimates of the strength of each vendor for each factor as the market progresses through its life cycle.
Time-Series Analysis – is used two ways. First, where the time scale is weekly, monthly or quarterly, and the objective is to identify and drive the model by decomposition into trend, seasonal and non-seasonal components. Second, as a preliminary data smoothing operation prior to the application of one of the other method.
Agent Based Models – are used to create forecasts by modeling the buying decision process at the individual buying unit (consumers or business units) level and then aggregating to the total market. Using cellular automation methods, large number of cells representing buying units are organized and linked to represent communication networks. Various economic, demographic, sociological and other influences that govern communication and buying behaviors are represented by parameters set for each cell.
Time is represented by an iterative process that evaluates the state of each cell in terms of both communication and purchasing behavior, in each time period. At the core of the model are stochastic functions that govern the probability that each cell will communicate and/or buy in that time period based on the value of the influencing parameters of that cell, as well as the state of the cells it is linked to. The model is run many times to produce not only a most likely forecast, but a distribution of probable alternative forecasts.
Trackers & Bottom-Up Models
Daniel Research Group also designs and develops Vendor Unit/Revenue Trackers for clients, enabling Bottom-Up modeling and forecasting, or for use in validating or calibrating Top-Down models.