A DISTRIBUTED ARTIFICIAL INTELLIGENCE SIMULATION OF BUDGETARY DECISION MAKING : SUMMARY OF FINDINGS

A study of budgetary decision-making in retired households based on personal interviews and computer simulations modelling spending and saving choices.

For more details contact:

Mr. Edmund Chattoe or Professor Nigel Gilbert, Department of Sociology, University of Surrey, Guildford, Surrey, GU2 5XH; Tel. +1483 300800; Fax. +1483 259356.

Key Points from the Research

The work of the project can be divided into five parts. The first part involved collecting data on budgetary decision-making in retired households using detailed interviews. The second part was to develop general hypotheses about budgeting from these data. The third part was to develop a framework for budgetary decision-making which showed the relationships between different aspects of the interview data and accommodated previous economic and sociological research. The fourth part was to use computer simulation (the representation of a social process as a model in the form of a computer program, rather than mathematical equations or a verbal description) to investigate the implications of the hypotheses and suggest where additional data or research were required. The final part of the research was to demonstrate that this process, beginning with interview data, developing hypotheses and investigating their implications with simulation (leading on to the collection of new data) is a useful way to proceed. Traditionally, economics has collected quantitative data on choices between goods, rather than decisions inspiring them, while sociology has used qualitative methods but only considered the budgeting process tangentially. In the past, most economists have declared both qualitative data and simulation inappropriate for the study of economic decision-making.

The table below describes the five levels of budgetary decision-making found to be important in the interviews. The levels differ in frequency of occurrence, impact and routinisation of response. For example, life course events occur seldom, have profound implications for subsequent choices and no appropriate routine response. Shopping choices occur repeatedly, with few implications for subsequent choices and have appropriate routine responses. The levels also differ in precedence. Decisions made at higher levels are treated as fixed, at least in the short term, with respect to decisions at lower levels. However, in the longer term, feedback from the lower levels influences the higher ones.

The table also explains the relationship between the current project and previous research. Economics focuses on shopping choice decisions and implies that this model of choice is appropriate at other levels too. Sociology concentrates on lifestyles which people pursue through their purchase of goods. This project attempts to link these two levels by investigating how the process of managing money facilitates a set of activities while shaping the process of shopping choice. (Further research is needed into decisions about life course events and contingencies.)

Decision Level Examples

Life course events:
Marriage, children, professional training, retirement, house purchase, incapacity

Contingency planning :
Taking out insurance, joining a pension plan, saving "for a rainy day"

Lifestyle planning:
Getting a part-time job, joining a drama group, planning a holiday, finding a lodger

Budgetary planning:
Saving 20 pounds weekly for a holiday, moving a bill to direct debit, cutting back on "luxuries"

Shopping choice:
Searching for a suitable coat, deciding between meat and fish for dinner, picking a tin of beans.

Several of our findings can also be understood in terms of this table. Because economic theory focuses on choices from a well defined set of alternatives with known preferences, it marginalises aspects of budgetary decision-making which are important in practice:

• Decisions about the allocation of time are made prior to decisions about the use of money. This is because well-being inheres in activities. (Shopping is just one activity among many.) Despite this priority, the link between activities and expenditure runs both ways and cannot be treated otherwise. Household members will often engage in part-time work to improve the balance between leisure and income and cutting out activities is a far more effective means of economising than buying more cheaply.

• Many goods (durables) persist through time rather than being consumed immediately. These are both an additional source of stability (retired households often have stocks of clothes and furniture) and a source of potential uncertainty (e.g. car breakdowns cannot be predicted). Economics suggests that purchase of durables can be separated from other sorts of purchase. In practice, almost all purchases display some durability which is also vital to the effective use of time, for example monthly shopping trips.

• The existence of uncertainty about the future makes choice based on well defined preferences and alternatives problematic. It is hard to compare the thought of a cake next week with the taste of a cake now. Where probabilities and utilities cannot be attached to future events because of uncertainty and incomparability, decision cannot involve optimisation and must be adaptive, relying on simple rules of thumb such as comparing what happened this week with what happened last week. (Rules of thumb are also useful when agreements need to be made between household members.)

• Unlike choices between known and separable alternatives, the allocation of time is highly constrained by the environment and its relationship with well-being is complex. People need to learn the contribution which particular activities make to overall well-being and this provides a further explanation for the observed stability of activities. (If too much is changing at once, it may be hard to say which changes have been for the better.)

• Finally, it is important to note that all these effects are interlinked. There has been a tendency to build models in which each effect is treated separately. This is both empirically unsupported and predictively unsatisfactory. Simulation is important in allowing interacting effects and processes to be modelled effectively.

In addition to the empirical findings described above, simulation has allowed us to investigate several other observations from the interviews:

• Households typically maintain at least two "accounts", with "rules" determining transfers between them. Sometimes there really are two bank accounts, or in other cases the distinction may be between a state pension paid in cash and an occupational pension paid into a bank. The "current" account deals with routine expenditure and the "savings" account deals with items which are expensive or otherwise unique. Part of the importance of the two accounts is to make an objective distinction between the two kinds of money. Money is rarely transferred back to the current account from savings. Instead it is spent directly, but only on certain sorts of "projects" such as holidays or replacing durables. The two account system is a powerful and simple method for balancing routine activities and uncertain events. Households can attain a balance by trial and error depending on their individual circumstances, considering only whether the balance in each account is increasing or decreasing.

• Learning from other people's budgeting behaviour is a significant influence on individual budgeting. The role of social learning has been largely neglected in economic theories of consumer behaviour, although in our simulations, social learning proved to have more effect than individualistic learning. Individualistic learning, that is rational calculation and learning from previous experience, is likely to be effective only in relatively stable environments where predictions about future income and expenditure can be made with a degree of certainty. If the environment is unstable, there are advantages in pooling many separately acquired experiences through imitation.

• Imitiation processes also have the potential to yield emergent effects such as the stratification observed in our simulations. Because simulated agents select which other agents to imitate, and these may reciprocate in turn, differences may become accentuated in a feedback loop which moves agents into relatively separate groups or clusters whose members share preferences for certain activities. In the simulation, this clustering was overlaid by income stratification. Income, although not directly observable by other agents in our model, did influence the patterns of their activities, simply because without adequate income, the agents could not afford to carry out some activities. This resulted in "rich" agents being able to benefit from relatively expensive activities which they imitated from their rich peers, while "poor" agents either did nothing, not being able to afford many of the activities in their activity schedules, or gradually populated their schedules with cheap activities copied from other poor agents. Over time, the rich and poor agents acquired almost entirely different activity profiles, corresponding to quite different "lifestyles".