Evidence of Intent

This post is a warning about UPS in New York. I missed a UPS delivery 3 times, called the 1800 number and was told to pick it up at 522 Greenwich in downtown Manhattan. The pickup center could not find my package, and was staffed by rude and unmotivated personnel. They were also looking for the package by hand, without any type of computer help. The UPS call center has no connection with the pickup center, you have to call the pickup center itself to confirm that your package is there.

By setting up their central call center this way, UPS shows a blatant disregard for the worth of customer time. The disparity between the delivery logistics (which are quite good) and the end-point exception handling (absolutely terrible) is amazing.

Econtalk this week mentioned an incredible fact, that the US GDP has a very tight growth band around 3%:

I was surprised that I’ve never known this before. The reason, I think, is that usually we see plots of year on year GDP growth, which look like this:

It is difficult to see that not only does the average (well, the geometric mean of 1+growth, to be exact) come to 3%, but also that the averages for shorter intervals of time also come to 3%; the deviations are highly anti-correlated and cancel themselves out fairly quickly, but this is not visually obvious.

Additional reading:

http://www.econbrowser.com/archives/2009/03/trend_stationar.html

http://econlog.econlib.org/archives/2009/03/greg_mankiw_get.html

The reductionist analytic modeling recipe goes like this:

1. Pick a system to study, choosing a boundary around a region in the world and mentally sectioning it off – everything inside that boundary is the system and everything outside is the environment.
2. Simplify the system by picking a few details to pay attention to, ignoring everything else. Divide the system into components with defined states and define interactions between components. Some parts have defined interactions that reach through the boundary defined in step 1, and interact with the environment.
3. Assemble those parts back into the whole, building up complexity by computing how the various combinations of states of the various parts interact with each other.

Step 2 simplifies while step 3 complexifies. Step 2 is semantic while step 3 is syntactical. If the results from step 3 successfully predict the behavior of the system, then the simplifications in step 2 are said to capture the essential details of the problem.

By thinking of the work accomplished by analysis as due to these three types of work, you could find the limiting reagent and spend more time on that. The analytic engine you are improving could be personal, collective, flow or even batch-oriented.

Your strength as a rationalist is your ability to be more confused by fiction than by reality.  If you are equally good at explaining any outcome, you have zero knowledge.

– Eliezer

Logic is equally valid in all universes, and hence cannot tell you which of those universes you are in. Applying logic does not give you new information, it merely aligns your decisions better to the information you already have, by either reducing bias or increasing the efficiency of your estimators.

It is good to distinguish when you are getting new information and when you are processing existing information better, because knowing explicitly how the validity of your ideas flows allows you to make corrections faster and easier.

Long tails in distributions are troublesome for 2 reasons:

1. They are hard to test for empirically because they represent rare events. How they look like in any given model is more model-driven than data-driven.
2. Models which are modular, and construct the distribution of interest from many independent component distributions, tend to underestimate long tails in the distribution of interest. This is a problem of degree, not a black and white issue – theoretical proofs use absolutely independent component distributions, and using those proofs for real work requires an assessment of whether components are independent enough in reality. That assessment is non-trivial and all too often skipped.

1. Forcing a concession
2. Convincing the other person
3. Maximally updating your own beliefs

“I don’t know which is worse… that everyone has his price, or that the price is always so low.”
– Hobbes

The capitalist framework is quite general, because as much as capitalist theories dictate when one should invest, they are completely agnostic as to what/when one should consume.

The idea of consumption is exogenous to capitalism and can be defined to be anything. For investment to take place, all that is required is a sufficiently low discount rate. For example, let’s say you want to save starving children in Africa. If you value feeding 10000 starving children in a year more than you value feeding 1000 starving children now, you have a 900% discount rate and should invest in anything with greater than 900% annual return. This particular hurdle rate would almost be impossible to overcome, but most altruistic goals have lower discount rates than that.

Extra credit: does morality have a discount rate?

On Monday’s Econtalk, Taleb said that knowledge was harmful. Here’s my take.

Formalisms are born in academia, where they establish boundaries within which solutions are sought. Those boundaries make it possible for ideas to engage, by directing people to compete on the same problems. Boundaries are necessary for focus, they make it easier to share and hybridize ideas. The real game (of life) is broken up into sub-games which academics play in for status and respect.

The sub-games of academia have independent  intellectual histories and traditions. Within each sub-game, there is enough competition that solutions are dependable and valid. However, to actually use an academic result in the real world, one has to make sure the sub-game itself is set up correctly. Sub-game incorrectness takes longer to catch – paradigms have inertia. Assumptions are often accepted by players to make the competition a good game with clear winners, not so much to reflect reality.

Alas, the academic quantification of risk in the finance sub-game has falsely enthralled us all, when in fact it is a weak shadow of real world uncertainty.

In decision theory, risk adversity is a terminal goal

In decision theory, risk adversity is represented by concavity in the utility function. Concave functions bend downwards, so a straight line drawn between any two points would lie below the function line, and averaging any two function values by any weight would have a smaller value than the function at the averaged position.

Px*f(x)+Py*f(y) < f(Px*x+Py*y)

For a given average outcome, the expected utility is higher if there are fewer possible outcomes. Maximizing expected utility then makes the agent prefer a smaller variance in outcomes. This is how decision theory generates risk adverse behavior.

Presented in this way, the desire to reduce variance is fundamental, an ends in itself, and not subject to justification.

Risk adversity seems to be instrumental in reality

When I consider a risky choice, however, I don’t undergo anything like the process of utility maximization. If I am highly uncertain, I may

1. delay the decision and acquire more information
2. avoid the choice to avoid complicating the context of future decisions
3. avoid the choice because of the fear that uncertainty makes me susceptible to fraud, since other agents have freedom in the uncertain space.

These reasons all point to uncertainty as an indicator of incomplete information, and as a cue to spend more on information or a chance to avoid future information cost, something a priori excluded if you assume information costs to be zero, as does canonical economic theory.

Presented in this way, the desire to reduce uncertainty is instrumental, a means of bettering the expected outcome, not a goal in itself.

Conclusion

Risk adversity’s relationship to the management of ignorance cannot be denied. Characterizing risk adversity as the maximization of concave utility functions misses this point. Exogeny is useful for compartmentalizing and communicating theories, but in making ignorance exogenous you restrict an agent from choosing to learn more or to avoid the cost of future learning. In the presence of this restriction, the model will not display a realistic reaction to risk.

Related

• http://en.wikipedia.org/wiki/Search_theory
  

Note: I use the words “value” and “goal” interchangeably in this post.

Terminal values are pursued for their own sake. Instrumental values are pursued as a means to attaining terminal values, or other instrumental values closer to the terminal values.

For example, if I had the ultimate goal of maximizing profit, I might break that up into the instrumental goals of increasing revenue and decreasing cost. Depending on the situation, those two instrumental goals can be broken up into even more specific instrumental sub-goals.

Instrumental sub-goals help to reduce repetition in thought. Conditions change in such a way as to change the actions demanded by an instrumental sub-goal, but not change the validity of the way in which the terminal goal is divided into sub-goals.

Irrationality is often caused by holding on to instrumental sub-goals even though they no longer serve your terminal goal. This is what Eliezer calls thought caching.