Unpredictability

Many decision makers spend enormous amounts of time and money trying to develop models to precisely predict or forecast the future state of a system. From a system dynamics point of view, however, this is a poor use of a decision maker's resources. There are two reasons for this. The first is that it is impossible, in principle, to precisely predict the future state of a nonlinear feedback system, except in the very short term. The second is that, even if it were possible to predict the future state of a nonlinear feedback system, a decision maker's resources are better spent trying to predict the behavior mode of a system in response to a proposed policy change, and in trying to redesign the stock-flow-feedback structure of a system so that it behaves well, regardless of what happens in the future. 

Any real system, whether it is a physical system, biological system, or social system, is continuously shocked or buffeted by external (exogenous) forces. An aircraft, for example is shocked by bursts of wind. The human body is shocked by changes in temperature (e.g., stepping into a cold shower). A firm is shocked by sudden changes in the demand for its product due to, say, changes in customer tastes and preferences. Figure 22 depicts a system that is being continuously shocked by external forces. 

When a decision maker attempts to forecast the future state of a system, he or she is essentially trying to forecast the timing, magnitude, and direction of the incoming shocks that are perturbing the system, so that he or she can respond to them. The implied mind-set of the decision maker is that the shocks, rather than the stock-flow-feedback structure of the system, are responsible for the behavior of the system. The decision maker's search for the causes of the system's problems is outward or towards the shocks, rather than inward or towards the stock-flow-feedback structure of the system. From a system dynamics point of view, the decision maker's resources would be better spent trying to redesign the stock-flow-feedback structure of the system so that it responds well to shocks, regardless of when they arrive, how large their magnitude, or in what direction they push the system. The system dynamics perspective is an inward or endogenous point of view. 

In order to illustrate how it is impossible, in principle, to forecast the future state of a nonlinear feedback system, consider the following experiment. Figure 23, below, is a simple system dynamics model depicting the interaction between elephants and hunters. In the experiment, this model is defined to be the "real world system." Next, an exact copy of the "real world system" is made. The "model" is perfect in the sense that its nonlinear stock-flow-feedback structure, its parameters, its distribution of random variates, and its initial values, are identical to those of the "real world system." The "model" is thus more perfectly specified than any actual social system model could ever be in the true real world. 

The experimental simulation of the "model" and "real world system" is set up so that, in period twenty, the "model" begins utilizing a random number stream that has been initiated by a seed value different from the one that initiated the random number stream that is being used by the "real world system". In other words, before period twenty, the "model" and "real world system" are completely identical. After period twenty, the "model" and "real world system" are identical in every way except for the seed values that initiate the streams of random numbers exciting their behaviors. 

Figure 24 is a time series plot of the simulated stock of hunters from both the "model" and "real world system." Before period twenty, the two curves are clearly identical and overlay perfectly. After period twenty, however, they begin to diverge significantly. Indeed, from about period thirty forward, the perfectly specified "model" of the "real world system" predicts the correct number of hunters in the system, only by chance. The conclusion is thus that, even a perfectly specified model cannot predict the future state of a nonlinear feedback system, except in the very short term. Point prediction is thus impossible in principle.