We want to predict the effect of small disturbances on the operation of a company in the market.
It is well known that the dynamics of a system can be examined based on its response to small perturbations. There are several assumptions here: The system does not change over time as a result of these small inputs, the system is at least approximately linear around its operating point, the system does not have a memory, and the input noise is under the control of the experimenter.
Taking the fourth point first – can we isolate the disturbance.
A large company has many disturbances impinging on it – meeting with analysts, release of quarterly figures, a lawsuit. Some disturbances take time to work through, some happen overnight. One needs to continuously handle many disturbances and "zero out" the effects of all the others to be able to extract useful information about any one of them.
What about if we look at the effects of disturbances on companies where not much happens – it is too large a leap to assume that busy companies or companies in the stockmarket’s eye will react in the same way.
The economist’s approach to this is "all other things being equal", but this is no help when there is no control over the input – and contrived experiments will often lead to the exact opposite response to that expected. If you keep making optimistic forecasts to boost a stock price, eventually no-one will believe you.
Companies are run by people, who are very good at reacting to disturbance and changing at least the apparent response. Something has caused the share price to go down – release an optimisitic estimate of future profits. The whole market is going down – release a pessimistic estimate so the spike is bigger when the market turns. The system is made up of many actors in the drama, each with their own memory.
A company lays off 5% of its workforce – the shares go up on the assumption that there will be more for investors. The company lays off 30% of its workforce, the shares go down, investors believing the market for the company’s products is collapsing.
The effects of two different disturbances can mask each other, effectively cancelling out, or can get out of phase, so the effect of each is magnified. This is particularly likely if one disturbance is used to cover the effects of another.
We need to find patterns in the swirling maelstrom that is the market. It is not impossible, just very hard.
We need to disentangle many effects, some with considerable lag. We need to bear in mind that market psychology is a fickle beast – the same disturbance leading to a rise in the value of a variable one day may lead to a fall another day, with seemingly no good reason (there is a reason, all those memories out there interacting with each other and with events they can’t control).
All that is saying is that there is no simple relation and a large number of variables need to be tracked to find worthwhile correlations.
We would suggest the use of a knowledge model – one that is making use of history, all those perturbations in the past, but is also capable of synthesising all the diverse effects into a prediction of the effects of a new disturbance in the current environment.
An optimistic forecast from one company when its competitors are forecasting gloom and doom – not believed, whether it is true or not. Another optimistic forecast, after previous forecasts from the same company proved false. The knowledge model needs a memory of what happened in the past, and what is happening now around it.
The knowledge model provides a framework for both analytic and experiential knowledge, and both are essential to making sense of the response of a complex system like a company.
By moving range information through its network, the model can make use of the analyses resulting from the work of many people, while rubbing off the hard edges from analyses of single effects done in isolation.