My First Agent Economics Model: Version 0.2.
Below you will find displayed the second installment of a demo of the agent
economics model that I am currently putting together.
This is a slightly more advanced demonstration of the model to test some more
ideas and and get a better idea of the internal workings of the reactions between
consumers and producers.
This Model adds the concept of individual producers to the previous milliEcon model, These producers, whilst still more
advanced than the some of the more basic concepts of producers, still lack some
of the strategic behavior that firms and producers can exhibit. A classic example
of this can be things such as over production and market share. The large balances
acquired by the monopoly firms during the start up phase (as the market collects
stock in order to manage market demand) should give them much more leverage
over the market than they currently exhibit, The fact that the link between
producers and employees is currently missing (its currently an 'open ended model
with no back propagation between firms and employees) means this model acts
more as a partial analysis than a solid overview of a simple closed economy.
What is it?
This model is the second version in the milliEcon series, it follows and utilises
basic economic principles to generate its data and display them in the graphs,
when the project is eventually completed, the overall design structure should
look something like this:
However, I'm not quite there yet... The UML diagram for the applet on this page
looks more like this:
Contrast this with the UML
model for version 0.1
If your wondering the reason for the map in the background of the goal UML
model its not there just to look pretty, it signifies that each individual element
is to be linked to a geographic location, this location plays an important part
for externalities and transaction costs associated with individual elements
interactions, in this case you will probably be interested in the game
of life-earth model which I implimented more as a distraction than for any
This model revolves around 3 'sets' of goods, if you wish, consider them as
'breakfast' 'lunch' and 'dinner'. Each good set consists of 5 individual goods,
you could consider these as ingredients. However the actual design interface
is not dependant on number of goods, firms or consumers, or even the types of
goods that are to be depicted, current developement work at this moment revolves
around providing a simple intuitive interface to allow definition of goods,
firms and consumer areas to the model, even integrating this into the TBM data
mining system which will allow the AI to do this automatically with minimum
Hopefully the interface should be fairly self explanatory, the two curves displayed
in the bottom left hand side represent the industry total cost and total revenue:-
all firms within a good set are homogenous. A slightly more detailed explanation
of the model basics can be found on the original milliEcon page.
You can also change firms cost using the drop down boxes in the table above
the cost/revenue graph.
On to play time:
Play with it, the box in the bottom right allows you to control the producer
elements of each of the 15 goods, such as the type of market structure (monopoly/perfect
competition etc) and also view some more production information such as any current
deviations from equilibrium in the market (such as over/under production) as
well as view some details of individual producers.
N.B. The applet requires the java
sun VM v1.5 or above installed on your system
If the applet doesn't start displaying data, refresh this page....
Who, why? and what next?
© Mark Parker, aka mSparks.
Please direct any comments and suggestions to my email, The graph display for this
applet model uses the java graph class library written by Dr Leigh Brookshaw
of USQ Austrailia Mathematics and Computing department You can get the source
for this applet (provided under the GPL) HERE
The agent economics concept, which is well documented by Leigh Tesfatsion
is a concept I have been working with since late 1998 (even if I only discovered
that recently), specifically within the Artificial Intelligence area, started
as a late night realisation of a fairly simple algorithm that potentially explains
the very fabric of every interaction we see in our day to day lives. however,
as with the extremely simple model you see above, that fairly simple algorithm
is quite far from simple, in fact, its more of a nightmare. However, gradually,
over the last 7 years, it has evolved into something of more substance than
the two simple pages of html it took to get it down and explain it. That, for
now, you dont get, but you can read more on how it fits with other AI systems/algorithms
in this article..
Most of the AI testing to date has taken place so far within the AQ2:LTKTBM
game, This model forms the foundations of the next generation of work that is
going into the TBM_AI SDK, LTBTBM - Licence To Breed The Borg Matrix. This game
itself is a fairly mammouth undertaking, but with the use of GPL tools, code
and the content provided by the increasing pool of supporters its quickly coming
into view beyond anything I ever imagined when I first wrote those fateful pages.
On top of all this, it has led to some most interesting possibilites with regards
to predicting market prices. Taking some simple assumptions with regards to
oil producers, published results regarding current details of oil consumption
and its relationship to other goods - along with some creative manipulation
of the graphs to align the time and past twelve month data the following 3 predictions
were made on the 01/11/05 Using data from the BBC market data pages.
Standard noise $3-$4, very similar to the variations required by the
market model to assess market response to price change.
start November to end December prices:- flat to fall then rise by a small
percentage, centered around $56-$57..
Beginning of 2006 - rapid upturn, overvalued through january/february
but peeking after several months of continual rise at a record ~$75 by
mid March, followed by a downturn to continue in mid 2006, but not returning
lower than $58-$59.
As of January 2006: These predictions have been further refined over the past
few months to predict the average weekly price for Brent Crude to within 50
cents for the week ahead. It should be noted however, that these predictions
are missing some important factors, which could greatly change the volitility
through the beginning of 2006, most notibly the effects from exchange rates
which have not been factored in (and the interest rate implicit within them)
and also certain shocks which could arise from demand shifts, currently idle
fields becoming active, and upcoming cost reductions such as pipelines, an interesting
point to watch will be how these pipelines affect oil company profits, a consistant
result coming from this model is that cost cutting can lead to vastly reduced
profits, unless the firm(s) benefiting from the reduced costs are very carefull
about how they alter production, or can expand this new production into previously
As of 23rd March 2006: In a follow up to these predications, all went well
up untill the 1st February 2006, when the data experienced an unexpected drop
of several dollars, while several factors could explain this, I think it is
most likely to be related to some change related to the handover of the new
US Federal Reserve chairman Ben Bernake from Alan Greenspan which occured on
this date. All in all a fairly strong result given the limited infomation these
predictions were produced under.
As of 23rd November 2006: Further analysis has revealed the close match of these
predictions with actual data, the expected peak of approximately $75 by mid March actually
hit this peak in May, and then remained overvalued until September, hitting another peak >$75
in August, and is currently hovering around the predicted $58-$59 mark for the end of the year.
All in all a successfull excercise, the next 12 months however look significantly less certain, mostly
due to an increase of potential standard noise to around $10-$20 for at least the next 4 months, expected volitility in interest and exchange
rate differentials between the various countries involved and the unknown potential impact of the significant change in the US balance of power. The course
of the next 3-4 years at least will be mapped out by events and prices over the coming months, the only reasonably consistant prediction among the various probablilities is
a peak of $90 sometime in 2007, the average trend however could be anywhere between slightly above that and somewhere around $40, depending on how that standard noise closes or opens over the coming months
LTBTBM is a huge world based simulator, which utilises distributed computing
and is designed to greatly enhance the access to data useable by the TBM AI,
this AI system is designed not only to run around maps killing the enemy, but
perform complicated risk assessment and inginuity on virtually everything thats
thrown at it - to start with this will probably just involve building up its
economics dataset and allowing the AI entities to interact with each other based
on these rules.
Not much is available for now, and theres still some way to go, but things are
moving at a healthy pace, some early screenshots (taken before I came to study
at Boston College) of the mapping technology it uses are available below, some
of the artifacts in these images still need fixing, but most were fixed before
I left - these images are all rendered in real time: