Decision Making Under Uncertainty

Other than principles of long term investing, the only other area where probably I have received good amount of information is on Decision Analysis (a.k.a. making long term irreversible decisions under uncertainty). The exposure to Decision Analysis was thanks to my stint in a boutique consulting firm which was on a cutting edge of decision analysis tools, theory and application. Interestingly, long term investing is also decision making under uncertainty and uses similar concepts but applied a bit differently. I am attempting to encapsulate the learning from the two field in such a way that it can make our day to day decision making better. Decision making under uncertainty is a complicated field and several books can be written (and several books have been written), however I will try to summarize it briefly.

Decision analysis can be done in a jiffy or can take months. It all depends upon how irreversible the decision is, what is at stake and how accurately you can calculate the likelihood of future events and their impacts. I believe for any important decision, it is worthwhile to do a quick & dirty decision analysis which mostly takes just a few minutes (and sometimes few seconds).

I. It is all about the decision tree

In essence, decision making under uncertainty is a very simple concept. Before taking any decision, you find out if things turn out your way, what is the value of upside you can have and if they don't turn out your way, what is the downside you can have. Then estimate the likelihood of things turning your way. You take the bet if:

1. Likelihood of things going your way * value of upside - likelihood of things not going your way * value of downside (otherwise known as expected value) looks extremely good to you. Note that in calculating the the expected value, don't include unlikely events (events that have less than 10% or 20% chances of happening).
2. You can ensure that there isn't even an unlikely case where the path will lead to ruins or a loss you can't recover from. If there is such a chance, you try to create backup, or share the risk with other players or you try to insure/mitigate the risk. If you can find no way to mitigate this unlikely but disastrous risk, you don't go down the path no matter how attractive. Obviously there are unlikely risks like a comet hitting the earth which you ignore entirely. But even a 1% chances of dying while trying to do a particular activity is too much. It just needs a developed intuition to see what kind of unlikely risks are significant in a particular case.
3. Unlikely upside potential should be ignored unless it can increase your investment by 100x, i.e. the upside potential is too huge compared to the chances of it happening.

Ultimately it is all about quickly estimating three things, (i) On an average will I gain or lose & by how much, (ii) Is there a chance that I will incur unacceptable damage which cannot be mitigated (iii) Is there a chance that it can be a game changing bet (while I won't lose much even if things don't go my way).

As an example, when there were just a few Coronavirus cases in the country but there were lockdowns in some other country, was it a good decision to stock up the house with months of supplies. If you ask the three questions (on an average in which case do you win and by how much, is there a chance of unacceptable risk, is there a chance of it being a game changing bet), the answers that I came up with were: (i) On an average you will be bit of a loser if you stock up since it is inconvenient to buy & store months of supplies, you end up spending large amount of cash upfront and you may even be seen as a panicky person. If number of cases are very few there is a significant chances that a lockdown may not come for months or maybe it won't ever happen , (ii) There is an unacceptable risk attached to not stocking up the supply. A small but non negligible risk did exist that there will be a lockdown earlier than you expect and stores may not have the supplies you need, (iii) There is no "game changing" possibility other than the key risk which we already identified earlier.

So what do you do? I stocked up on March 16 when there were a little more than 100 cases in a country of 1.3 billion. Though in most cases, I "lose" a bit but I avoided an unacceptable risk. In the end it turned out that stocking supplies made things slightly more convenient but one could have done without it. But it was always nice to sleep peacefully. If something didn't happen it doesn't mean it couldn't have happened.

II. How do you find out how likely something is?

Trying to estimate likelihood or probability of something happening is extremely difficult and many experts say that this shouldn't be done unless there is sufficient data available to back up your estimation. There typically are three scenarios in which the likelihood of various outcomes can be estimated:

1. When you have a large enough past data-set which can be used to calculate probabilities of future events: New drug discovery and oil exploration are two areas where decision analysis works like a charm. The reason is that the past data can be used to reasonably estimate the future probabilities of a drug becoming successful (given test results) or an oil well being productive (given exploration data). It is a no-brainer to use decision analysis in such scenarios.
2. When you can find a good enough cause-effect relationship: When investing for long term, it is impossible to accurately find probability of any particular outcome. But we use reliable cause-effect relationships. If we can reasonably estimate that somebody will keep requiring a product (customer segment & their need is going to exist), the target company has a sustainable competitive advantage which is evident from numerous similar historic cases, it has a good control on margins and the cost to grow the business is low, we can reliably say that the cashflow of the company is going to increase in future. It is impossible to assign an exact probability to it, but it is sufficient to know it is much more likely than not. Certain trends which may be hard to notice over a short term but compound over time can also add to the confidence. These include rising income, productivity growth etc. Actually I believe long term investing is all about betting on these so called "small" trends which become gigantic over time.
3. When the probabilities are truly difficult to assess: There are certain cases where the probabilities are truly difficult to assess for you. For example, you do not have enough competency to determine the cause-effect relationships, or there are so many random factors affecting the outcome that even cause-effect relationships cannot give you a good idea. In such cases it is best to avoid a decision altogether if you can. Try to fish in some other pond. For example, a long term investor doesn't try to predict stock prices (at least in short term) since too many random factors can affect the outcome. But in cases you cannot avoid making a decision, trust expert assessment. Here when I say experts, I am not talking about people who confidently predict the future. Except for rare cases, a very confident prediction about uncertain things generally means the person is likely to be biased or ignorant. Try to find a person with a successful track record in that area but who is open to admit what he or she doesn't know.

III. How do you find out what the likely upside or downside can be? Especially what are the unlikely but disastrous cases?

This is something which is extremely tricky. People have an innate need to feel certain about the future. That is why their sub-conscious brain prevents them into going to the true extent of good case as well as bad case. My theory is that the true variability of our future is so unnerving and evocative (in many cases unnecessarily so) for most of us that our subconscious brain tries to protect us from it.

Here is a typical example to highlight this human weakness. When working with a software company I asked 3-4 senior team members of the product development team (all of them brilliant software engineers from IITs) about the best and worst case of the time needed to finish the current product they are working on. I tried to sensitize them on all possible uncertainties, issues that they have not thought about but may crop up, different modules may not integrate, the requirements may change etc. The worst case scenario of development time given by any member to me was 6 months. The software finally came out after 18 months!!

From what I have learnt, the first step to find the best and worst cases is to find the underlying factors most affecting the outcome. In most cases there are 3-6 factors which can explain most of the variation in outcomes (and don't try to use decision analysis where this isn't the case). It is also important to understand how those factors are linked together. The linkage helps explain if some of them are likely to become adverse together (which can make potential outcome much worse, as a classic example job security and stock market can become adverse at the same time and so it is important to have liquid investments outside of stock market).

It is also important to find the second and third order impact of the decision to find out potential outcome. For example, the outcome of getting a good job is not the salary you are getting, but the entire career you are getting which can be many times more rewarding than the initial salary. Similarly, if you let a corrupt person stay in your company, you will be inviting more corruption in future. These are examples of second order effects that we typically overlook

Then you find out the best and worst case of these underlying factors (better to base it on past records rather than you intuition since as explained earlier human intuition is almost always wrong when trying to predict best and worst case). This will give you a much better idea of how well or how bad things can go in future.

While it looks unnerving, for most common situations you have to deal with only 3-4 variables and it is fairly easy to predict the possible outcome through this process once it becomes a habit.

IV. How much time should you invest in creating the decision tree, calculating the likelihood and the value of upside & downside?

As already stated, it all depends upon how irreversible the decision is, what is at stake and how accurately you can calculate the likelihood of future events and their impacts. If it is truly a life changing irreversible decision it makes sense to even use excel. If you can accurately calculate the probabilities and outcomes based on past data, it may make sense to have a model in place. In most other important situation a 1-2 sheets of paper and a pen are sufficient. For most daily situations, a minute of thinking may suffice.

V. Finally, be aware of the common mistakes

The common mistakes which I am aware of include:

1. Not choosing a good frame: Frame refers to the scope of decision. As a rule, frame should include the entire canvas on which the decision is going to have an impact. For example a frame can be: "Should we build the new factory" or "What can we do to meet the growing customer demand and preserve our competitive advantage". In most situation the second frame is better compared to the first.
2. Not considering all options: ?One of the other human weakness is also not to be able to see all options. For important decisions it generally requires to ask the question to multiple people with different backgrounds to come up with the options and reject the options only after serious thoughts. You need to avoid the situation of a bunch of carpenters who can think of only one option, use the hammer.
3. Not considering the "intangibles": Our love for neat equations and single answers (blame the Principia Mathematica) many time leads us to ignore the intangible factors. If we are trying to find which job is better for us, simply projecting the salary to next 30-40 years is insufficient (though it is important to have an idea about it before deciding). Perhaps more important factors would be whether you will love the job, can you adjust to the environment or are your strengths aligned to the key requirements of the career. Because if those three things are not affirmative maybe you will not have those 40 years or you will have miserable 40 years despite the financial upside of the career.
4. Not collecting enough data: In my experience, the number one mistake smart people make is that they do not collect enough data. You need to pick as many brains, read as much as you can and get as much data as you can before deciding on really important irreversible decisions.
5. Not appreciating the inherent human incapability to assess probabilities and predict extreme cases: You can take my word for it. The best case will always be much better than you had imagined and the worst case will be much worse. Things that you never thought will happen are going to happen. It is always better to use wider estimates than what your brain tells you and insure against completely unpredictable event. Our brain switches off to uncertainty because it thinks we are not prepared. Being prepared just makes it easier for it to navigate uncertainty better.