There are a couple of books which I think , every risk manager should read. After reading them I hypothesize that there will be 4 outcomes
- Risk manager ignores the view as just another view and moves along ( and moves the firm too , in a downward direction)
- Risk manager starts introspecting his/her methods and starts becoming skeptical about them and will probably do something about it
- Risk manager starts introspecting his/her methods and becomes extremely skeptical about them and will do something about it, hopefully
- Risk manager quits his job and does something else, far removed from the field of risk management. Who knows, he might just embrace risk with out giving any consideration to the frame of thinking at his/her previous job
"Plight of the Fortune Tellers" is one such book which should be on the must-read list for any risk management professional. It is a book which was written years ago , before this subprime mess hit the economy. Its insightful view of the risk management world from the world of frequentist and bayesian worlds is bound to shatter a lot of myths for an open-minded reader. I always intuitively knew that bayesian is the way to go. But this book reinforced my thoughts as the author brings a lot of facets and goes very deep in to the subject to show that Risk management's current practices are completely flawed. BTW, there is not a single equation in the book . It is a remarkable achievement for any author on risk management to avoid equations and explain his ideas in simple english!!. This book rocks!
"The Cardsharps", by Caravaggio ,
shows a handsome and probably very naive young man who is engaging in a financially rather risky activity:gambling. We become aware of how risky this game of cards really is when we note that a man is looking over the shoulders of the young player and communicating to his fellow cardsharp the cards in the young man’s hands: unfortunately, a lowly “three.”And, when our attention is drawn by the action of the playing cardsharp to the dagger concealed in the waist band, we begin to fear that what the young man is risking may be more than a few ducats.
The dire consequences in miscalculating the odds of a risky activity, is what the painting talks about!
Here are the main ideas from the book:
Author pitching to the reader to spend time and effort on the book
- Leibniz usage of statistics to forward the cause of Prince Fredrick of Prussia. Statistics is used to take decisions and not mere hypothesis about stuff and write papers about it
- Fundamental need to rethink about the role of probability, role of rare events, time homogeneity and forecast period!
Thinking about Risk
- There are 2 types of probabilistic assessment. System I and System II
- System II is analytical, tools, math based
- System I is heuristics based, where we overestimate odds of one group and underestimate odds of another group
- Risk management in the current environment is heavily skewed towards System II thinking
- What is needed is a good blend of System I and System II thinking
Thinking about Probabilities
- Frequentist Probabilities and Subjective Probabilities
- Subjective does not mean "anything goes". You should be able to bet with those odds
- Why cant we say that coin is 40% fair. Why should we always live in the world of accept/reject world ? Either coin is fair or biased ?
- Point estimate + confidence intervals == frequentist world...
- Risk management needs Bayesian Probability where there is a constant interaction of prior assumption and data
- With limited amount of data, in the frequentist world, there is always an external model which one uses!
Making Choices
- Prospect theory and Utility theory - What have they got to say on choices ?
- Problems associated with each of the above theories
What is Risk Management for ?
- In a bank or a firm, whats the purpose of the risk management division. Shouldn't investors themselves indulge in diversification and any form of risk management done by the firm will not suit all the investors ?
- The truth of the matter is that investors almost always do not have the complete information about what is happening in the firm
- All the investors see is stock price and earnings , their volatility. High earnings volatility is a signal to the investors that the firm might not do well
- The sole purpose of risk management is to create stable earnings estimate. - This is what the author thinks. If you ask Taleb, he would probably say, firm should get rid of risk management division completely :)
Var&Co - How it all started ?
- 3-6-3 World of a banker
- Firms started to undercut the bank by directly issuing bonds
- Banks started banking! on trading revenues
- Complex instruments --> More desks --> More variables
- VaR , a convenient metric but which is actually VariableAndwRong
Looking beneath the surface?
- There are approaches to Var , namely Historical simulation, Empirical fitting, Fundamental Fitting
- Historical simulation is dumb but least controversial as it involves no parametric calibration
- Empirical fitting is intellectual orgasm. Pick one of the distributions in the stats101 book , find the best fit and then you have all the freedom in the world to talk about 99.979999 percentile!!!(arggg). Even if you have data about 1000 days, you can still predict a 10000 day event. Does it scare you?
- Fundamental fitting - This is what happens when physicists try see finance as a science. Finance is social science and there are no universal structural equations!! alas!! its not pondered upon a lot
- Monte Carlo - More sin has been committed with Monte Carlo technique than any other, in the statistical world. WHY ? Montecarlo doesnt say a thing about distribution. Given a distribution , it will tell you how to proceed...But ...who gives you that initial distribution
- Central Limit theorem - It doesn't work in tails. It talks only about the initial moments!!! So, rare event is out of question using CLT
- Correlation is a metric for elliptical joint distributions. But why should joint distributions be elliptical for the observed real world variables ?
- If I want to assign meaningful probabilities to very rare events, I cannot escape the link between the frequency of data collection, the relevance of the data, and the rarity of the event in question. And if the event I am interested in depends on the co dependence among very many variables, the complexity of the problem literally explodes
Which probability matters in risk management ?
- Probability Square - Four vertices - Frequency of Data Collection + Time homogeneity + Rarity of the event + Forecast Horizon
- Frequentist world : High frequency of data is good as the event is time invariant.
- We need external model /view which says that the data observed in the past is relevant to the forecast horizon. But then how do we validate that external model ?? It is not math / sciences where one can take axioms as god given and move along
- If the phenomenon I want to investigate is very time homogeneous, if I can collect data very frequently, if what I am interested in happens not too rarely, and if the horizon of my statistical prediction is short, I am clearly in Frequentist Land
- Prediction of Short term horizons - Market Risk - Large data sets available ; Is it relevant data ? For valuing CDS how far backward is relevant data ?
- Prediction of Long term horizons - Does market to market make sense ? Important to model trend , but it is inherently more variable than estimating volatility. Long term nature of the issue at hand cuts down the data points, cuts down the confidence.
- Statistics 101 - Point estimate + confidence intervals is crap in the risk management world
- In Markowitz’s Portfolio Selection bible, for instance, the possibility that all the probabilities attaching to the different outcomes may not be perfectly known only makes its appearance on p. 257 of a 303-page book—by which time the assumption that the objective probabilities are perfectly known has become solidly embedded in the psyche of the reader
- The frequency with which we can update the value of a long-term portfolio (if we do not mark it to market), the length of the projection horizon of interest for these portfolios, the difficulty in estimating the trend and its high dependence on our models and prior beliefs, the importance of the trend itself in determining the overall portfolio performance—all these factors strongly suggest that a subjective-probability approach is the appropriate one.
The promise of economic capital ?
- Need for the regulatory capital should be linked in a beneficial manner to the risk management of the firm. A bank should not end up gaming the regulatory capital requirements
What can we do instead ?
- Subjective probability is the relevant type of probability for assessing returns, and judiciously employed frequentist analysis is useful to assess risk, it then follows that we cannot use the same statistical tools to evaluate the risk and expected reward from a project.
- Treat Risk and Return separately and use different tools for each of them
- For risk, look at the asymmetry of the distribution, dispersion and kurtosis
- Three decisional crutches - Best Possible Scenario Worst possible scenario , Break even analysis
- A call to arms for related disciplines such as bayesian analysis, decision theory and experimental psychology
This book cannot be speed read at all. It needs to be read slowly, allow the points to sink in and ultimately help you understand the nature of risks in a better way. Personally, after reading this book, I think there is tons of innovations that need to come in to Risk management field. WHY ? The current mess is a clear indication that things are not right and risk management as a discipline should strive harder than ever before to prove its worth.
