Scott Ashton and Dr. Teppo Felin, Organizational Leadership and Strategy
When I applied for an ORCA grant one year ago, the stated purpose of my study was to “design, develop, and write the code for an agent based computer simulation which will explicitly attempt to model the social processes by which nascent firms and their entrepreneurial founders are matched with workers to thereby create an organization.” Once the simulation was completed, I hoped to be able to publish the results in a joint publication with my faculty mentor. Of these two goals – first creating a simulation, and then publishing the results – I have successfully accomplished the first but was unable to accomplish the second this year. Despite disappointment at not being able to publish, the skills I gained while working on the project (more than 150 hours of programming and researching) have opened up exciting new avenues for research with other professors and have also connected me with some of the premier researchers in business strategy at places such as Harvard and U Penn.
The rest of my report will proceed as follows: First I will describe how my mentor and I decided to change our question from describing the processes of organizational formation to studying competitive dynamics between firms. In the second part I will describe the evolution of our model as it underwent three distinct iterations. In the final part I will describe the opportunities my research has opened up to me as well as questions for further research.
Soon after I had submitted my ORCA proposal, my mentor expressed his interest in switching our research from studying organization formation to studying competitive dynamics, a somewhat related question. His inspiration was a highly cited 1989 paper which described how profitable companies aren’t necessarily found in favorable industries with high markups etc, but can be found anywhere where future profitability is not well predicted by existing firms. If the other firms do not expect the strategy to be profitable (although it in fact is) the firm that predicts accurately will reap the lion’s share of the profit, while the other firms are competing fiercely in the same market space. The research question we are interested in is why some entrepreneurs and the teams they assemble are so accurate at predicting market opportunities and why others are less so. My professor wanted to simulate and understand the competitive dynamics of these entrepreneurial teams with different abilities to predict the future before he moved on to studying how they were formed.
The first computer simulation we created was fairly simple in concept and implementation. To make the simulation easy to picture in your mind, first imagine a large circular O sliced into many blocks with each block representing a unique industry. The distance between blocks represents how separate the two industries are – for instance, on one end of the circle you might have a block representing the steel industry and on the other end there would be a block representing high end pet grooming services. Next assume that each industry has a different size – for instance, steel will be a multi-billion industry while pet grooming will only be in the millions. The size of the industry was randomly assigned to any given spot. Next imagine some arbitrary number of firms competing on this board. Whenever two or more firms are located on the same industry, the pie is split equally among them. The object of the firms is to accumulate as much money as possible, which they try to do by moving from industry to industry. However, the firms face two limitations. First, they can only look a certain distance to their right or left when trying to decide where to go. Second, each firm is given an “intelligence” which determines how accurately they are able to perceive the profitability of each industry and the number of competitors also there. Once these game mechanics were programmed, I ran several hundred simulations and found that the model wasn’t very useful. There was a slight correlation between firm “intelligence” and profitability, but that was only to be expected. In short, the model didn’t give us any counterintuitive insights, which we thought might be due to its simplicity.
The next model was similar to the first one, although more complicated. First we modified the world from being circular to multi-dimensional, so that firms couldn’t only move right and left but forwards and backwards and so on. Think of it like a rugged mountain landscape. Each unique point on the landscape represents a different industry, or company strategy, with the height representing the size of the industry. Additionally, whereas our first circular board had randomly determined payoffs for each industry, in the new landscape we could tweak the “ruggedness” to make it either smooth or jagged. These modifications still failed however to make the model very useful. Inevitably the firms would begin clumping on the same industries as each of the firms began heading towards the highest peak they could see. But when the firms all arrived, the profit-pie would be divided amongst all of them, making the industry very unprofitable. Then, the firms would move as a herd from one place to another, never satisfied.
For the next model, I thought to circumvent the clumping problem by developing a “gravity model” of competition. In this type of model, the payoffs weren’t evenly split between firms sharing the same industry space, but rather having lots of firms in the same general area of the map decreased the payoffs. Imagine a bunch of bowling balls fighting on a trampoline – the number of firms in an area determined how far the payoff was decreased. The final modification we made in this model was to make it possible to further tweak the ruggedness of the landscape. In a couple small ways, we also simplified the model so that’s it’s extremely similar to the ones being used by current researchers in the top journals. However, the model is still not quite done. Even my newest modifications have not been able to get around the clumping problem and I probably will need to change the gravity model to something a little simpler. My current advisor has recommended one potential solution which shows some promise. If we can find a model which does a decent job of recreating what we see in real life, I’ll think we’ll have contributed.
To conclude my report, the more than 150 hours I spent working on this project have been extremely valuable. I learned many programming skills which I never would have learned otherwise. I was able to gain insight into the way research is conducted in my discipline. The best thing about this project is that it opened up a door for me to work with another strategy professor at BYU, David Bryce, for whom I have made an interactive applet which lets humans instead of programs make decisions in virtual worlds similar to the ones I’ve described. This applet has been forwarded onto Daniel Levinthal at the University of Pennsylvania and Jan Rivkin at Harvard University, both of whom are strategy chairperson’s at their respective universities. For my honor’s thesis, I hope to create an interactive experiment with my professor and build on the research I conducted in all of these simulations.