This post is about how you can have a scientific approach to thinking or analyzing different phenomena using Bayesian Network. The only tools I used to prepare information were OpenOffice's Spreadsheet and the tools I have provided for this blog. I was looking for some data source on the web and finally found the Canadian to US dollars exchange Rate in last nine years in Bank of Canada website. If you just draw the rate for 2007 to 2015, you'll have something like the following graph.
|Canadian to US dollar exchange rate (data source: Bank of Canada)|
|up / down / no change state of Canadian dollar exchange rate|
|Bayesian Network which shows the relation of the quarter|
of a year and Canadian dollar exchange rate changes
The graph has four quarter node named 1,2,3 and 4 and like the previous one three rate change symbols "up", "down" and "nchngd" (for not-changed) Now we have more information to decide what the next day's rate will be and it depends on what quarter we are in at the moment. If you want to see when the rate has most ups, look at all the arrows end to "up" node, you see in quarter two the chance of having more "up" is more than the other quarters. Or for downs, quarter one is the worst.
|Bayesian Network which shows the relation of the month |
of a year and Canadian Dollar Exchange Rate changes
OK, now if you write a program to build these Bayesian Networks and then based on the given training data try to estimate the next 30 days, for som runs of the application you'll have the following trends.
|Trend estimation for Canadian dollar based on information in this post.|
Here I have supposed in the first day we have a rate of 70 (for 0.70), and then the colors show every different runs the Bayesian Network gives us. Now you can have some analysis of these trends to have a better understanding of this crash position. Note that this model doesn't tell you why it is happening, it just helps you to see how you can do some analysis with this basic information to improve the way you look to the future.
So as you see, it is unlikely to have a better rate in 30 days, while it is more likely to have it lower than 0.70 cents.
This post is not the correct place to talk about what happens when we continuously use the Bayesian Network or if we need to update the network as we derived some new data, or how much error we have after driving a new estimation, etc. We will talk about them later.