What is a Model?
A good model is a simplified but accurate representation of reality that can help us to better understand reality and to develop effective ways (strategies) of interacting with reality in order to reach our goals. All people automatically build models of the world around them, and of themselves in relation to the world, in order to function effectively in the world. Models are by definition simplifications, so they necessarily leave out certain aspects of reality. Good models leave out aspects of reality that don't have significance or relevance for the function or behavior we are trying to model. Bad models leave out aspects of reality that are significant/relevant and therefore, are not accurate representations of reality. If your model of reality is wrong, any strategy that you build based on such a model will eventually fail. This is an important point so let me repeat it:
IF YOUR MODEL OF REALITY IS WRONG, ANY STRATEGY THAT YOU BUILD BASED ON SUCH A MODEL WILL EVENTUALLY FAIL!
An Example of a Model
The above explanation can seem a bit abstract and that's because it is. In order to get a better understanding of the concept of a model, I'm going to present an example of something almost all of us have many models for: CARS.
Although we never explicitly state it, most of us “know” what a car is and have many models in our brains that allow us to identify and use cars to reach our goals. For example, most of us have a model of a container on wheels, usually four, that moves people and stuff, at a speed between 0-100mph, from point A to point B.
There are other details that we include in our models about cars. For example, we know that most cars have 4 doors and that some cars have 2 doors. We know that cars have a steering wheel which is used to turn the front 2 wheels left or right. We know hundreds if not thousands of little details about cars that we have accumulated throughout the years based on our experiences with cars.
It is important to recognize that although we all “know” what a car is, everyone has slightly different models for the category “car.” A car mechanic will know much more about the components of a car and the function of those components, because a mechanic's job/goal is to fix those parts when they break or malfunction. On the other hand, a layperson might only know that a car has an engine, an exhaust pipe and a place under the hood to refill the windshield wiper fluid. The layperson's goals for owning a car are to get from point A to point B successfully and perhaps to impress their friends with the brand of the car that they own (status signaling).
The key takeaway from the above example is that we build models to help us reach specific goals. If our goal is relatively simple, such as driving from one town to another, the models we need in order to reach our goals will probably be relatively simple. If our goals are more complicated, such as repairing or building a car, then we will need more complicated models to reach such goals.
Building a Model
If you want to build a model to represent a process, you first need to collect data. Direct observation is the most accurate form of data collection. Unfortunately, this method is slow, inefficient and often not an option. For example, if you want to study the price series of Microsoft Corporation common stock from the year 1990-2002, you have to rely on historical databases that collected this information during that time period. If the historical stock price data you obtain gives you only the daily prices for the stock, that is one single closing price for the stock per day, it will be impossible to use this data to build a strategy that trades intraday (multiple times per day). The key point here is that data will naturally limit the quality of the model you will be able to build which will limit the types of strategies you will be able to construct around it.
As an example of data collection, imagine there's a new employee at your office and you want to develop a way to predict whether they will be late to work on a specific day. Without any knowledge/data, your prediction rate should be no better than 50%, since they can either be late or on time. If you carefully observe them over a six month period and spot a pattern in their arrival times to work, you can use this new data to create a better prediction model. Perhaps you might notice that they are usually late on Thursdays because they take the train that day instead of driving. You can exploit such a trend by betting that they will be late every Thursday and on time every other day. The more relevant data you collect, the more you can refine your prediction model and formulate better predictions.
After collecting the data, you need to confirm the quality of the data to make sure that it accurately represents the phenomena you are trying to study and model. You also need to define the limitations of the data to have an idea of what you can realistically expect from this data set. Let's take our Microsoft Corp common stock (MSFT) price series data from above as an example. You might go to Yahoo Finance and download the closing prices for MSFT. But how do you know that this data is accurate? In this case, you would want to collect multiple data sets from multiple independent sources for the time period you are studying and compare those data sets using statistical tools. You might find that 1 out of the 5 data sets has aberrant prices after 03/29/1999. Upon further investigation, you might find out that MSFT had a stock split precisely on that day. Perhaps the source that compiled that particular data set didn't reprice the stock price after the split. During the course of analysis you will have to make many important decisions. For example, do you throw this one particular aberrant data set out or do you normalize the prices to reflect the other data sets? You have to be able to justify and keep track of the various changes made to the data set because these changes can cause inconsistencies down the line.
The next step in building a model is data analysis. Once you’ve collected the data and prepared the data, you need to assign meaning to the data by analyzing it for patterns. If the data has no patterns and is seemingly random, you will not be able to build any meaningful strategy from such a data set. Analysis is not easy. It requires a basic understanding of statistics and in some cases, understanding and the use of higher level math. Analysis also requires intuition and a general understanding of the field or phenomena that you are studying.
Building a Strategy
So now that you've built a general model and teased out some patterns, you can begin to construct a strategy that will take advantage of the patterns in your model. The types of patterns found in your model will often imply the types of strategies you should construct. For example, our fellow coworker who is usually late on Thursdays provides a semi-predictable pattern that we can construct a strategy around.
Let's say you and a number of your fellow coworkers decide to create a betting pool to bet on whether this employee will be late or on time. As described in the previous example, you might develop a strategy of betting LATE on Thursdays and ON TIME on Monday, Tuesday, Wednesday and Friday. But there is another component to the strategy which is the frequency and size of bets placed. Let's say that your total capital for betting is $100. Should you bet $100 every time? Unless you are 100% certain that you will be right, and this is never the case, you should never bet your entire capital base on a single event. Let me repeat that because it might save you from bankruptcy one day:
UNLESS YOU ARE 100% CERTAIN THAT YOU WILL BE RIGHT, AND THIS IS NEVER THE CASE, YOU SHOULD NEVER BET YOUR ENTIRE CAPITAL BASE ON A SINGLE EVENT.
So how much should you bet? This is a somewhat complicated question to answer. If you are betting on 5 different events, that is whether the employee will be on time or late for every day of the week, you should at least split your total betting capital into 5. You would at most want to bet $20 per day. But even this might be a bit too large of a bet size. Even though you’ve spotted a pattern in your data, i.e the employee is usually late on Thursdays and usually on time on all the other days, it would be helpful for you to collect information about the relative frequencies of those outcomes. Since you’ve observed the employee’s behavior over a six month period, this gives you roughly 24 observations for each weekday. Let’s say that the employee was late 19 out of the 24 Thursdays, or 79.16% of the time. You can expect to win your Thursday bet at about that rate. But there’s another problem. What if the employee is on time in streaks? Perhaps they were on time for 5 Thursdays straight. If you had bet on them being late for 5 straight Thursdays, you would have exhausted your entire betting capital if you were betting $20 per event. So perhaps you want to be on the safer side and bet $10 per event. You can see that this can get very complicated, very quickly. This is the world of strategy development. And we haven’t even discussed what you stand to win on every bet. Maybe you only stand to win $2 for every $20 that you bet. Is it even worth participating in this betting pool with this type of payout?
The reality is that the above example doesn’t give us enough statistically significant information/data to make intelligent decisions about developing a winning strategy. Although office betting pools are fun, they will not bring you consistent profits, so don’t quit your day job just yet.
In this blog I’ve defined what a model is and what a strategy is and laid some basic principles for the development of each one. Of course the real process for doing this is much more complicated and can include many other factors. Because markets are complicated and highly competitive, models and strategies need to be constantly reanalyzed and modified to factor in new developments. After all, patterns in markets appear and disappear as participants take advantage of those patterns to make money.
I offer training/education in market structure, model and strategy development and trading. If you are highly motivated and willing to do the hard work to learn these skills, you can contact me for an introductory session.