Technology Forecasting Methods

In general, technology forecasting methods can be classified into quantitative and qualitative methods. Martino (1993) outlines ten forecasting methods, including Delphi, analogy, growth curves, trend extrapolation, correlation methods, causal models, probabilities methods, environmental monitoring, combining forecasts, and normative methods. The following section will separates these methods into quantitative and qualitative categories and discuss their contents and applications. Furthermore, the criteria of selecting forecasting models are also presented.

2.1.1 Qualitative forecasting methods

Qualitative forecasting methods are those forecasting methods do not use mathematical or statistical technologies, therefore, based on this definition, Delphi, analogy, environmental monitoring, and normative methods.

1. Delphi

Delphi is a kind of expert opinion methods and a series of questionnaire are used to collect the panelists’ opinion. Panelists are the experts familiar with the specific industries or problems that the research focuses on. Delphi is usually applied in the three conditions. The first condition is there is no historical data can be used; the second condition is when important external factors happen and the previous data can not be applied, and the third condition is when ethical or moral considerations dominate the development of technology (Martino, 1993). Delphi can gain the extensive opinions and panelists do not need to interact with each other, so they won’t be influenced by others and shift their opinions. Levary & Han (1995) suggest that all participants should be

experts about the given technology.

2. Analogy

The assumption of analogy is if the background or characteristics of two technologies are similar, they may have the same similar development trends; therefore, the forecast can be based on historical analogy. There are nine dimensions need to be taken into consideration when applying analogy, and they are technological dimension, economic dimension, managerial dimension, political dimension, social dimension, cultural dimension, intellectual dimension, religious-ethical dimension, and ecological dimension.

3. Environmental monitoring

A breakthrough of a technology is the end result of a chain and is not easy to predict. Environmental monitoring method then can be applied to detect the breakthrough and is a systematic forecast method that involves evaluating kinds of environmental sectors, including the technological sector, the economic sector, the managerial sector, the political sector, the social sector, the cultural sector, the intellectual sector, the religious-ethical sector, and the ecological sector (Martino, 1993).

4. Normative methods

Normative methods are the goal oriented forecast methods and are different from the exploration forecasts which are used to predict future development using the previous or present data. Setting a goal is the first task of applying the methods, and therefore, when, who, what, why, and the know-how of a technology should be clearly set. The representative models of normative methods are relevance trees, morphological models, and mission flow diagrams.

2.1.2 Quantitative forecasting methods

When a forecasting method uses mathematical or statistical data to forecast, the method is a kind of quantitative forecasting methods, such as growth curve, trend extrapolation, correlation method, causal models, probabilities methods, and combining forecasts. Valid time series datasets are the key to forecasting accuracy of quantitative methods and forecasters believe that the historical data present a logical trend for future and can be used to project the prospect development (Martino, 2003).

1. Growth curve model

Growth curve model, also names as S-curve model, is often applied to forecast the technology or product lifecycles since the growth of technology/product lifecycle is usually follows an S-shape curve. Growth curve model uses the historical date to forecast the future performance of the technology or product. Three assumptions need to be fulfilled before applying the traditional growth curve models. The first assumption is the upper limit to the growth curve is known; the second assumption is the chosen growth curve to be fitted to the historical data is the correct one, and the third assumption is the historical data gives the coefficients of the chosen growth curve formula correctly (Martino, 1993). Since the growth curve model is the main method of this study, the detailed discussion will be presented in the next section.

2. Trend extrapolation

Every technology will reach its maximum performance level, i.e. the upper limit, and new technology will appear to replace the old one. Trend extrapolation is used to forecast the progress beyond the upper limit of the existing technology. Even we do not know what and when the breakthrough technology will be, the previous technology development and historical data can be used to forecast the future technology. However,

if a technology is known that it will not have further development, the trend extrapolation is not suitable to be used. Levary & Han (1995) suggest every trend extrapolation model need to have assumptions and satisfying these assumptions is the determinate of forecast accuracy.

3. Correlation method

Correlation method is similar to analogy method. The predicted technology should have the similar characteristics to the previous technology. However, correlation method uses the quantitative historical data of the similar technology to forecast and analogy method uses qualitative dimension to project the similar technology development. Martino (1993) introduces several correlation methods for forecasting a technology, including a technological precursor, cumulative production, total capacity, and economic factors.

4. Causal models

Causal models are used to realize the reasons that induced the development of the technology. Once the reasons are defined, the future development can be forecasted.

Martino (1993) introduces three types of causal models. The first type is technology-only models, such as the growth of scientific knowledge and a universal growth curve, and these models assume that the technological changes can be explained by internal factors of the system of technology. The second type is techno-economic models and the assumption of these models is that the technological development is caused by economic factors. The third type of causal models is economic and social models which assume economic and social factors are the reasons that induce the technological development, and KSIM and differential equations models are the major representative models.

5. Probabilities methods

Probabilistic methods are used to predict the range that a technology can develop and reach to and the probability distribution over the range. Martino (1993) outlines two types of the probabilistic forecasts. The first type of methods relates a range of possible future values and the probability distribution over the range and the second type of probabilistic method is based on a probability distribution of the factors that produce technological changes. Probabilistic forecasts can be operated using simulation techniques.

6. Combining forecasts

Every forecast method has its advantages and disadvantages, and therefore, different methods can be combined to improve the accuracy of prediction. Combining forecasts then are popular in predictors to avoid problems of selecting only one forecast method. Researchers should study the strengths and weakness of individual forecast methods to know how to combine different forecasts to reach better predictions. Usually combining forecasts can be quantitative and/or qualitative methods. Trend and growth curves combination, trend and analogy, components and aggregates, cross-impact models, and scenario analysis are most used combining methods. Levary & Han (1995) suggest the cross-impact analysis should be applied when the factors that affect the future technology are known, and the scenario developers should expert all aspect of the technology.

2.1.3 The selection of technological forecasting models

Levary and Han (1995) outline six factors that influence the selection of forecasting methods. First, money available for development, the more money invest in the given technology, the more opportunity the technology can be realized and the

shorter the development time. Second, data availability, what data researchers can retrieve affect the methods selection. Third, data validity also affects the choice. Fourth, uncertainty surrounding the success of technological development, some technology forecasting methods are suitable to high uncertainty situation while other are not. Fifth, if similarity of proposed and existing technologies is high, analogy or correlation methods can be applied. Finally, number of variables affecting the development of technology, the more influence factors, the more complex models should be applied, and therefore, combining forecast methods may be needed.

Young (1993) applies nine growth models to determine the procedure of selecting an appropriate growth models. The author concludes that the most important procedure is to identify the characteristics of datasets before fitting the data into the growth curve models. Predictors intend to apply growth curve model need to know the knowledge of upper limit, to observe whether the fifty percent takeover point has been achieved in the dataset, and to study the length of datasets.