Economic Fitness Landscape
A fitness landscape – also called an adaptive landscape – is a three-dimensional state space used to describe the environment within which adaptive agents operate. It is an important approach used for modeling complex adaptive systems of all kind.1 An economic fitness landscape is this type of model applied to understanding and interpreting the macro dynamics within an economic system or industry. From the perspective of complexity economics, a macro economy is a complex adaptive system, that is to say, that it is composed of many interconnected autonomous parts that are capable of adaptation.2 Other examples of complex adaptive systems include flocks of birds, ant colonies, the human immune system, the Internet’s routing system, cities and all forms of social organization, from financial markets to political regimes.
Complex adaptive systems are highly dynamic. They are typically high-energy systems. They import a lot of energy in order to maintain a dynamic state far- from-equilibrium. In physics, this is what is called a dissipative system.3 They import energy and dissipate it in order to maintain this dynamic non-equilibrium state. A flock of birds would be an example of this. Because the components have a high degree of autonomy and are capable of adaptation, this means that control is primarily distributed out to the local level. The overall state of the system is a product of the interactions between the agents. Agents are acting and reacting to each other’s behavior, like businesses competing in a market or traders buying and selling in a financial network. Out of all these actions and reactions, we get the overall state of the system.4
We can model these interactions in game theoretical terms. An agent is involved in interactions of both competition and cooperation with other agents typically within its local vicinity, but also possibly globally throughout the system. Thus, an agent is involved in both zero-sum games of competition and some positive-sum games of cooperation, and may be involved in both at the same time, what is call Coopetition. For example, some companies share and collaborate in research and development projects but then compete in the market.5
Within this environment, agents are trying to improve their functionality or payoff according to some set of metrics. This may be called the fitness function; a simple parameter that defines how efficient that agent is at intercepting and transforming the resources available within the system. For example, this might be how much work does a company get and how good is it at turning it into net revenue, or how competitive is a country within the global economy, or a trader’s strategy within a market, or the risk-return ratio on a portfolio. All of these could be defined according to a single metric and compared across the different elements in the system, this single metric we would call the fitness function. Likewise, we might model the system according to a second set of parameters as to what function an agent within the system performs. For example, we would model economies according to industrial sectors where companies that perform similar functions are grouped into the same industry.6
Once we have this set of parameters, we can use them to create a 3-dimensional model, a kind of state space, that is to say, any single point in this 3-dimensional space will be defined as a state to one of the agents. The two axis that define the horizontal plane will capture the agent’s function. Agents that are in proximity on this plane will then perform similar functions. This might be businesses in a similar industry or countries with similar economic profiles or financial institutions that operate in similar markets. The agent’s elevation within the space will then be defined by the fitness parameter. The higher they are up on the vertical axis the more effective they are, and thus the higher their payoff.7
Thus we have a full landscape and agents are now placed in this landscape; they are able to adapt and they are trying to move up to higher elevations with higher payoffs. On the most basic level, agents face a dichotomy in their choices. They can either exploit their current position or they can invest resources in order to try and find some more optimal payoff, that is to say, a higher elevation. For example, a person can stay in their current job, with all of its security of salary, or invest their time and resources in finding a new one that might be better, or a business can stay exploiting its core competencies giving out high dividends to shareholders or invest in moving into new markets. Of course, agents may do the two in parallel but if we want to come up with a single value we can simply subtract the resources used for consumption versus those used for investment to get a gross positive or negative value.8
The strategies and choices that agents make will be dependent upon the type of landscape they are acting upon. Different economic environments will represent fundamentally different topologies and thus different strategies. The topology to the landscape will be defined by the type of system we are dealing with; starting with a very simple linear system and going to a very complex dynamic system. In an isolated linear system – without interconnections and interdependency between components – the topology will be very simple, a single dominant peak in the center of the topology representing the fact that there is one single optimal equilibrium. An example of this might be a hierarchical organization where there is one clear optimal position within the entire organization. A monopolistic market might be another example. There is one clear well-defined optimal position within the landscape. As long as we are dealing with a closed system and there are no interdependencies, it means that the landscape is not going to change. Given the example of a hierarchical organization, an agent can then adopt a very simple rule – stay trying to get promoted upwards until you get to the top and then stay there.
Now we can turn up the complexity by allowing for many agents with many interactions and interdependencies between them. When we have many different interdependencies, an agent’s payoff will be dependent upon the interactions between a number of different parts. Take for example a competitive market. How well a given organization performs in the market will be dependent upon many other actors. Because there are many interacting parts, there will be many different optimal solutions giving our topology a rough form with many different peaks, with any one of these many peaks representing a particular combination of all the organizations in the market. Some will be better than others but there will always be many different local peaks representing all the different combinations we might have.9
If we go a step further and allow the agents in the market to adapt, then as the agents adapt and change their state, this will change the payoffs for other agents, with the net result being that the whole landscape will start to move up and down. This is a lot more representative of a real-world complex adaptive system, like a competitive market, where a business faces a number of different competitors who are all altering their behavior and strategies over time.
Lastly, if we want to try and capture the true complexity of a market we would have to recognize that this whole macro system does not exist in isolation. It is, of course, embedded within a particular social, environmental, technological context, all of which are providing the system with a set of input values that define the overall landscape.9 When one of these input values changes, for example when a new technology comes along or a change within the political regulatory environment, then the whole landscape will change. This can, of course, be minor changes that are happening on a continuous basis, or major and abrupt transformations. It might be a major transformation in the political environment, such as that experienced by the Russian economy with the fall of the communist regime, or the current major technological disruption brought about by the rise of the Internet. During such a time it is no longer so much the interactions between the agents that define the topology, but now these input values to the system, as the agents become subject to the process of evolution that is acting on all the agents in the system as the whole environment changes.9
The strategy that agents adopt will depend fundamentally on the environment they are acting within. As mentioned, within a very simple environment that is closed, static without interdependencies, the agents only need a very simple rule telling them to stay going upwards. As an example, we might think about the former communist economy of Russia where the government would tell a company to simply produce as much of a product as possible with as little resources as possible. Within such a context you do not have to worry about competitors or the landscape changing, and thus there is no need for adaptation. All that is needed is a very simple optimization algorithm with a single equilibrium.10
If the environment involves many interdependent parts – such as companies trying to optimize the stock and transportation routes within its supply chain – then the problem becomes more complex. It requires significant investment of intelligence and computation. The agents need to stay trying different solutions, that is to say exploring the landscape going up and down on different local peaks, but gradually over time reducing the number of resources spent exploring – this is a very simplified explanation of what is called a simulated annealing algorithm. Because everything is held static, an agent can afford to make this long-term investment of resources trying to find one of the optimal solutions as it will be able to, then exploit this for some time in the future due to the static nature of the problem.10
Within topologies that are moving up and down due to different agents acting, reacting and adapting to each others behavior, agents have to stay constantly balancing exploration with exploitation. That is to say, unlike our previous example where after some time the agents settled into a stable state, within these more complex environments agents have to stay adapting indefinitely. If they stop exploring, they will slowly fall behind as others exploit the new opportunities. Likewise, the landscape will never really settle down because as soon as it does approach stasis, this will create an opportunity for some agent to act and then again some other agent to act as the whole topology becomes re-animated.
Lastly, when the change is systemic and coming from outside of the system – with little that agents can do to affect this environment – then the agents are no longer competing with each other but this is now a form of evolution. The whole environment is changing and performing selection on them. They have to be able to adapt to that environmental change.11 This requires a whole different level of adaptation. It is no longer a question of just responding to immediate changes, but the agents may need to change their whole functionality in order to maintain their fitness. For example, the advent of the Internet has brought systemic change to the publishing industry. Newspapers are now less competing with each other and are more competing to maintain relevance within this new environment. In order to do this, they need to be able to reinvent themselves on a whole new platform, under a whole new set of rules, as the whole environment has changed. This does not just test their capacity to adapt to a normal market environment because this will not be sufficient to reinvent the whole organization, instead, this evolution in the industry tests their degree of diversity. It is only out of maintaining some pool of diversity that an organization is best suited to developing new solutions in response to systemic changes within its environment.
From this, we should note that the transition from an agent or organization operating in a simple environment to a complex environment should map onto this parameter of exploring and exploiting. In simple environments, agents are primarily exploiting through optimization. In complex environments, they are spending more time investing and survival is always some interplay between both. Exploring and exploiting are very different activities that require a very different type of regulation and management. As the venture capitalist and writer Peter Cohan12 puts it: “The exploit business focuses on cost and profit, spurs efficiency improvement and incremental innovation, is strong at operations, has a formal structure, controls for margin improvement and productivity, values efficiency, quality, and customers, and leads in a top-down manner. By contrast, the explore unit focuses on innovation and growth, spurs new products and breakthrough innovation, is strong at entrepreneurship, has a loose, adaptive structure, controls for milestones and growth, values risk-taking, speed, and experimentation, and leads in a visionary and involved manner.”
From this, it should be clear also that how we try to regulate an economic system will change fundamentally depending on the complexity of the system and environment we are dealing with. Within a very simple context, we can use simple top-down regulatory systems with basic linear optimization algorithms. However, when the environment becomes more complex, these top-down regulatory systems may not be the best solution, as components need autonomy to stay adapting locally and we increasingly need to maintain and develop diversity in order to ensure sustainability. In such a context, more distributed forms of regulation are better suited. A core challenge for a large organization like a macro economy or corporation will be in how to integrate these very different paradigms in regulation and be able to switch between them.