Modelling the Process of Market Emergence

5 Results: the first simulation set-up

We have run the set-up for 73 periods. What we get is not too dissimilar to what we observe. Employment, for example, does not decrease steadily through the simulation. It does not show any dramatic changes at all although it has not been stable reflecting changes in the demand for the labour force following variations in production activities over the whole early period of simulation. The time path of employment is given in figure 2.

Figure 2: Total Employment

In production we observed a collapse after an initial surge (due probably to initial simulation conditions) but starting from date 7 production trends varied from sector to sector. The production of corn remained remarkably stable because corn has at least one relatively stable source of demand in the form of households. The production of both spades and iron shows considerable oscillation but it develops accordingly with the picture of unit sales which demonstrates certain cyclical pattern. Series for production is shown in figure 3 and for unit sales in figure 4.

Figure 3: Sectorial Production

Figure 4: Unit Sales

Total inter-enterprise debt increased dramatically over the course of the simulation run. The time series is reproduced here in figure 5. Though it lacks the slight upward trend of Figure 1, it is not badly out of line with what we observe.

Figure 5: Debts

We also find that wage payments are constrained by cash holdings of employers and that this restricts consumption which is, in turn, a main source of restraint on the economy as a whole. The number of firms which cannot pay their full wages bills is increasing in the classic pattern of an epidemic.

The results are not too bad but they could be better representations of what we observe. We could see preferred trading arrangements and groupings emerge. There could be more serious upward pressure on prices and rather more upward pressure on debts. We could determine the sectoral pattern of output changes and develop our simulation set-up better to replicate that pattern. But how should we seek to achieve these aims?

Figure 6: Paasche Price Index

Our simulation set-up comprises both a representation of the technological environment and a representation of agents' modelling language. It does not represent either a government or a foreign sector. Our main concern in this paper is to demonstrate the virtues of our representation of agent learning. These other aspects of the application domain are naturally taken up as we extend the present set-up.

Our experience in developing simulation set-ups is the most important effort in developing the representation of the agents' modelling activities. The results reported here, however, are based on the assumption that the agents have an extremely limited modelling language which did not enable them to specialize and generalize their models as appropriate. Two examples of agent's models are given in table 2 and table 3. The symbols beginning with `?' are symbolic variables. In this case they take numerical values but the values could be (and in other settings usually are) non-numerical symbols, lists or logical clauses. The first of these models simply says that if the last order value from farm-5 was larger than the current order value, then predict an increased value of sales. Such a model would imply that the input ordered from farm-5 is a constraint on the output of farm-13. Purchasing more from farm-5 would therefore allow greater sales by farm-13. The second model makes the same statement about the relationship between the cash holdings of farm-13 and its orders from farm-5. In this case, the relationships are well confirmed since they have been observed for three periods.

Table 2: The models of farm-13 at period 19 (a)

Table 3: The models of farm-13 at period 19 (a)

This is about as extensive as the modelling language got. Agents can recognize direct and in which they are confirmed and disconfirmed. Since even these limits change because of the multiple dependence of variable values, such models are easily and usually disconfirmed.

Generated with CERN WebMaker