Michael Baker - Thesis - Problems in Longterm Forecasting and Planning
About | Contents
Previous: 8. Planning | This: 9. Conclusions and Further Work | Next: A1. Sources of Data on Freight Transport
In this final chapter I shall give my views on the place of forecasting within the planning process, why forecasts are seen as being necessary, how they can be avoided and some insights I have gained into welfare economics. I shall end the chapter with some areas in which I think further work would be fruitful.
Two different views of the role of forecasting within the planning process can be held. In one view forecasting plays a central role by providing a background against which the formulation of programmes of action can be planned. It is this view which was put forward in the previous chapter. In the other view forecasting can be seen as coming after the point at which the general outlines of programmes of action have been decided upon. Forecasts are then used to fill in the details of these programme, for example by helping to decide what is "economic", and to define them in public. In their own ways both views can provide insights into the planning process.
My view of forecasting, as given in Chapter 7, have implications for the planning process. At the political or goal setting level the acknowledgement is required that forecasts are subjective, and at the practical of programme formulation level ways of avoiding forecasts or overcoming their deficiencies need to be found.
It appears that there is, at present, a very widely held belief by planning practitioners that forecasts form a value free input to the planning process. In a critique of traffic forecasts, and road transport planning, Adams (1980) points to the following circularity. In the absence of a clear policy, traffic forecasts are based upon the hypothetical government policy that the greater use of cars is desirable and upon an assumption that roads will be built to accommodate them. The resultant forecasts are then used to say that roads are necessary to accommodate the forecast increase in traffic. This is a good example of a forecast being used as a goal setting avoidance technique.
In "Social Forecasting: Predicting the Future or making History?" Miles and Irvine (1979)
have the following to say about the use of forecasting.
Adequate analysis of the conditions needed to establish a better
future is part of the process of creating that future.
Methods of assessing the contributions that alternative actions ...
might make to realising social goals are necessary, and some of the
existing forecasting techniques may well be applicable - if it were
absolutely clear that the use of these methods was itself part of the
process of making history.
Forecasting practices which treat people as doing no more than reacting
passively to changes beyond their control need to be replaced by people
assuming responsibility for democratically formulating and planning
social goals.
The implication is that forecasting is not a value free objective activity. However the adoption of Miles' and Irvine's suggestion would require new structures or modes of organisation within society.
At the practical level it is often acknowledged that forecasts cannot be made, however it is usually then claimed that something must be done and this is usually seen as being to make and use a forecast anyway. My view is that the solution to the problem is not to be found in improved forecasting technique but rather in ways of avoiding or reducing the need for forecasts. I shall say more about this below.
In the previous chapter I outlined some of the reasons why forecasts are required within the planning process. In the terms of the human mind analogy forecasts fulfil the requirement for a future image of the environment against which actions are taken. The length of time over which a future image is required is determined by the time lags in the system between an action being initiated and the perception of changes in the environment.
Under the view that forecasting comes after the basics of an action programme have been decided, forecasts are necessary to defend the actions decided upon and to determine the details of the programme. For example forecasts of increasing road traffic have been used in much this way both to justify motorway schemes and to make decisions as to which motorways to build. Another example is the use of energy forecasts to defend the current nuclear power building programme.
There are several ways in which the necessity for making longterm forecasts can or could be avoided. These include having shorter lead times, letting more functions within the planning organisation be carried out in automatic mode, introducing more feedback and accepting less "efficiency" within the system being planned. These four approaches can be directly related to the human mind analogy of planning presented in the previous chapter. Shorter lead times will principally have an impact upon the types of programmes which are developed and accepting less "efficiency" will alter the goals the system is working towards. In practice the distinction between more feedback and shorter lead times is rather blurred.
One of the problems inherent in many of today's large systems are long lags between an action being initiated and a result being perceived in the environment. There would seem to be several advantages to be gained from reducing these lead times in planning systems.
An example of this would be a change from the use of large generating sets in the electricity system to the use of smaller sets. These advantages would stem from the shorter length of time between ordering and installation. This would be beneficial because the shorter term forecasts required are more likely to be "accurate", and recovery from any errors made due to poor forecasts would be quicker.
Another way of avoiding forecasts is to move towards self governing systems and away from the external control of systems. This could be achieved by letting more units within the control system operate in automatic mode (automatic programmes). It would probably require the introduction of some of the other measures I have mentioned.
For example the introduction of a capacity tax on the price of electrical appliances (as proposed by the Energy Research Group, 1976) could be used to substantially reduce the necessity for making electricity demand forecasts. However this would need to be in conjunction with a move towards the construction of electricity generating plant which has shorter construction lead times so that the plant building programmes could respond quickly to changes in the capacity tax levied.
I would like to think that the human mind analogy of the planning process could be stretched to the point where automatic programmes could develop of their own accord once awareness was focused on the issue at hand. Such methods of developing automatic programmes would be similar to the methods of learning propounded within the "Inner Game" (Gallwey 1974, Gallwey and Kriegel 1977).
Many instances of self regulating systems can be found in nature. Two of their main characteristics are diversity and redundancy. With these, such systems are able to cope with large changes in their environment without the system collapsing. However such characteristics are currently not deemed to be desirable in the so called advanced societies because it is not "economic". For a given expenditure of resources more can be achieved if there is less diversity and redundancy. However the resultant systems tend to be much less stable to unexpected events. It is in such systems that forecasts of the changing environment are perceived as being necessary.
In conventional economic terms what I am suggesting is that a higher value or price should be placed upon system stability than is done at present. The direct use of a price would be very difficult to achieve in practice since there is no market for system stability and its price would be difficult to determine. However economics is not the final arbiter of what is done in society. There tend to be several overriding arguments to do with "insurance" and "national security". For example the army is uneconomic.
My views on ways to avoid forecasts (as above) are to some extent held by some others, for example:
The management of complex organisations is now recognised to be a continuous steering process, and hence planning to be concerned with the definition of objectives, i.e. what is to be achieved, the control of progress along it. Likewise, those concerned with governing and managing now accept that uncertainty is the corollary of complexity, and have therefore come to interpret anticipation in a somewhat different way. The planning process can explore alternative futures. But prediction and control of the future have come to be less important relative to anticipation of continual surprise and the creation of an organisation capable of responding creatively to changing situations. (Center for Environmental Studies 1973).
Commenting to this Cross (1975) says
In such an approach to the planning process forecasting is much less
important because of the short time horizons and because more rapid problem perception and speed of response is substituted for it.
However I don't believe that the general recognition that CES claims to
exist does actually exist within many planning environments such as central government.
Such recognition seems to be restricted to some academic circles.
Welfare economics is concerned with equity and efficiency. I have previously found (Baker 1974) that it cannot satisfactorily cope with equity because of the impossibility of making objective interpersonal comparisons. There are several standard defences of ignoring equity such as that there are mechanisms within society that ensure that the distribution of income is socially acceptable. Using such arguments welfare economics has tended to concentrate almost exclusively on the efficiency or optimal distribution of resources aspect. This has been because it is perceived that it is possible to do this in an "objective" manner.
I have now come to the conclusion that this side of Welfare economics (the efficient allocation of resources) is equally built on shifting sands. The foundation for most work on the efficient allocation of resources is the Pareto optimum. However one of the principal pre-conditions for the attainment of a Pareto optimum is perfect information and this can never be attained in the real world since there can be no perfect information about the future.
There are several different areas in which I can see an opportunity for further work as a result of this thesis. Briefly these are into:
As noted above (Section 9.4) I have now become aware that both areas of welfare economics (equity and efficiency) are built on shaky foundations. It is principally in the area of the concept of economic efficiency and the use of the Pareto Optimality criteria that I think fruitful work could be done. On Pareto optimality the main thing which needs to be investigated are the consequences of the non-applicability of the preconditions for the attainment of the optimum allocation of resources in the real world.
In particular I have concluded that objective forecasts are not possible and so it is not possible to have perfect information about the future.
As an aside it is interesting to consider the age old paradox of free will and pre-destination. In the following plan and planning could equally well be replaced by the optimal allocation of resources.
If free will exists it is not possible to forecast,
and so it is not possible to plan.If predestination exists it is possible to forecast,
but planning is futile.So, depending on whether free will exists or not,
planning is either futile or not possible.
Notwithstanding the conclusion that I have come to that objective forecasting is not possible and has no place in the planning process, and also that forecasting models are of their nature prone to error, I do see a need for more work on forecasting methods. However any such work should start from the acknowledgement that objective forecasting is not possible.
A large area in which further work could be done on forecasting methods is in gaining greater understanding of the ways in which the systems being forecast work. For example in the case of freight transport the first step might be to construct time series of volume of production, movement ratios and average length of haul for all freight broken down by commodity and mode as outlined in Chapter 1. This could then be followed by an examination of the underlying causes for the trends in these time series.
Of the three sources of projected values for input to forecasting models only trend projection and goals are autonomous inputs to the totality of the forecasting process. As shown in Chapter 5 (Figure 5.5) other forecasts are not autonomous. For example most forecasts of transport demand and of energy demand use population forecasts as one of their inputs. In their turn population forecasts use forecasts of economic growth, and so on.
It would be very interesting to make a collection of forecasts and trace the flow of forecast outputs used as inputs in other forecasting exercises. At one level this would entail identifying numbers input to a forecasting process which were explicitly acknowledged to come from another forecast. At a different and more difficult level it would entail identifying how individual forecasts and the general background of forecasts influenced the forecaster in making his judgements.
Above I have put forward ways in which forecasting could either be avoided or its role could be reduced. However there are many details which will have to be worked out if these ideas are to be implemented. These include both detailed studies of individual cases and studies of the underlying principles.
In particular planning situations examination will have to be made of which of the suggested approaches or mix of approaches to the avoidance of forecasting are most appropriate and how they could be implemented.
A major study in its own right would be an examination of how diversity and redundancy as prerequisites for robustness in systems could be incorporated into economics or alternatively how economics could be replaced by a new discipline whose results could be used to help in policy formulation.
Another study which would help to determine the cost of uncertainty would be an empirical study of how the accuracy of past forecasts has been a function of the length of the forecast time period. I suspect that a reasonable hypothesis could be put forward that the "accuracy" of forecasts is an exponential function of the time period over which they are made.
Long lifetime products (such as roads and power stations) are planned for due to a desire that they be used "efficiently" over their lifetimes. If forecasting problems mean that this is not possible perhaps the rules of the game (economics) should be rewritten so that the future is taken into account when decisions are made but that past signals (prices) are not carried into the future (by capital charges). This would mean that all capital formation would be out of current spending not future spending.
Another area in which work could be done would be examination of the feasibility of letting automatic programmes develop by the focusing of awareness in the appropriate place as mentioned in 9.3 above.
Within the control system model of planning (as described in Chapter 8) it is possible to identify systems which will not work as intended. For example consider the directives (or control signals) from government to a nationalized industry, such as the electricity industry. If the directives change at the same or faster rate as that at which the industry can respond to these changes, then effectively the government will not have the control it desires over that industry. The effect will be that the industry will operate at some time average of the control signals. If the environment changes faster than the control system can respond, then the control system will not be in control.
The control system model could form a useful framework within which to analyse the way in which planning is conducted. It could be used to see if the system can actually respond to the control which is desired over it. It could also be used to devise ways in which control systems will work as desired.
I now realise that my concern is not with forecasts themselves, but with systems with long lead times between initiating action and reacting to perceived errors. If the lead times are similar to the reaction time of the control system then oscillation may occur about a desired reference level. If the lead times are also long the cause of such oscillations may not be apparent.
I think that it would be fruitful to examine how systems with such long lead times arose and how the lead times could be reduced rather than worry about the problems of making forecasts. One of the ways in which lead times could be reduced would be to reduce the number of levels in the hierarchy of control at which reference levels need to change.
In writing this thesis I have had difficulty in giving "objective" definitions to the terms I have used. Often I have only had a vague idea of the meaning I wish certain words to convey, much like Gregory Bateson's "stuff" which is to be re-examined in more detail at a later date (Bateson 1975).
I have a feeling that the pressures to get a clear understanding of concepts before committing things to words (either verbal or written) may well have as much to do with the short comings of the language we use as with the desire or necessity to avoid woolly thinking. Maybe my perceived difficulty in attaining great clarity of thought is due more to the structure of the language I use to express my thoughts than the nature of clear thought.
These difficulties may be a product of the subject-verb-object structure of our language or its underlying metaphysics (of cause and effect). I think it would be well worthwhile examining the ways in which the structure of language structures and limits our thought processes. For example the necessity to distinguish something from everything else (i.e. to divide reality into separate parts) before it can be given a name gives rise to the concept of "objective" reality. That is something which exists independently of me. If the language structure did not lead to the "objective reality" concept then much of the confusion that I have been trying to address in this thesis about "objective" forecasting would not arise.
There may be some answers to this in Language Thought and Reality: selected writings of Benjamin Lee Whorf (1956).
About | Contents
Previous: 8. Planning | This: 9. Conclusions and Further Work | Next: A1. Sources of Data on Freight Transport
Copyright © Michael Baker 1981,2005. All Rights Reserved.