November 2009



This book makes a case that talent is something else besides the nature and nurture stuff. Author based on his field research says that there are three ingredient’s to talent which are
  1. Deep Practice
  2. Ignition
  3. Master Coaching


Deep practice involves struggling in a targeted way so that the learning happens at an accelerated pace. It is not usual “practice, practice , practice” mantra BUT one needs to stretch to reach a specific target and keep learning in the process. Roger Federer,for example does not play 1000’s of shots each day. He practices for a few hours and works on his timing. So, deep practice at the outset is a little different where one can describe it as error induced learning. It involves three stages as described in the book. Chunking the task + Repeating it + Learning to feel it.

The author gives a great blend of examples to support his case.

  • Bronte’s sisters little books
  • Brazilian football Futsal
  • Z Boys – Empty Swimming Pool
  • Florentine artists – Apprenticeship
  • Russian Tennis stars – Practicing with imaginary ball
  • Dallas Music School – How the students learn stuff differently

Ignition , the second attribute of talent is basically a mix of long term commitment, deliberate practice, persistence etc. The beauty of this book is the range of examples quoted like KIPPS Schools in US, South Korean Woman Golfers, Anna Kournikova’s role in Russia’s tennis talent upsurge, Scrooge Principle etc.

The third attribute is Master coaching. This aspect is a great read for a wannabe teacher or for that matter anybody who wants to groom talent, be it in a school / organization / game.

Applying the above stuff : Let’s take a programming language and let’s say you want to be good at it. How can one deep practice it ?

Writing 1000 programs each having a random objective would be helpful but the learning pace would be slow. Let’s say one wants to master data visualization skills. Using a specific library to carry out a task is ok but if you want to master it, you have got to create some specific mini projects and work towards them. These mini projects should be challenging enough to make you stretch , but at the same time, should be with in a reachable distance , else one might give up. So one of the challenges in mastering a language is to create mini projects which are with in one’s reach, where there is a vague idea that you will commit errors in the process of completing the mini project. These errors are the key to learning. This mini project could span a few days, to few weeks to a month. If one has not made any mistake for more than a week or so, then it is clear that one has set an easy target and the whole purpose of working on a mini project becomes futile. Now this has to be on a regular basis. So, one needs to be pretty creative to come up a list of progressively tougher mini projects which will grow your Myelin .What’s Myelin..well, that’s the stuff mentioned in the book which actually gives some scientific reason behind author’s arguments. Statements rooted in science are easily believable …In that sense, this book has a very strong case for an approach towards talent development.

Takeaway from this book filled with beautiful examples : Stop + Struggle + Make Errors + Learn is the key to developing any talent

Well, if not for anything else, the examples in the book make it a worthy read.



The only reason I chose to read this book on a weekend was to verify the null hypothesis of the author, ” an intuitive guide” 🙂 Well, intuition and econometrics are kind of animals that don’t sleep together that well in the SAME book. Either econometrics books are very math/probability oriented OR they are “stats for dummies” types. I had tons of other things to do on this weekend, but took a peek in to this ~100 page double line spacing book,just out of curiosity.

Points that grabbed my attention, not becoz of the math , but becoz of the nice angrezi that was used 🙂

  • Model building means moving from a unconditional expectation to conditional expectation.
  • Expected value of a Unbiased estimator equals the parameter to be estimated, irrespective of the number of trials
  • While unbiased is difficult to verify in most situations, Consistent estimator is what is generally needed
  • Examining the asymptotic behavior of the estimate is a clue to its consistency
  • OLS regression method essentially minimizes the variance of the error term by choosing the appropriate parameters for the linear model
  • Don’t ask whether random walk fits the data or not ? Instead ask , “How can I beat the random walk ?”
  • “Are mid-sized cars equally expensive in Japan and in the US, after correction for the exchange rate between these two countries ?” IS FAR BETTER A QUESTION than “Does the purchasing power parity hold good for Japan and USA? “
  • Economic theories deal with aggregated quantities..Empirical work needs fine grained data
  • 7 case studies to illustrate various models
    1. convergence of countries attributes – a Convergence model with no stochastic behavior
    2. Direct mailing problem – probit model for selecting customers for direct mail (a truncated regression model )
    3. Does automated trading improve trading efficiency – use multiple equation regression model for futures and spot
    4. Unemployment and recession linkage — auto regressive model with latent variable…. With all this jazz could anyone have predicted the 10% unemployment rate that the US is facing now !! Sometimes it makes me feel that all macroeconomic variable prediction is just some jazz and has no practical value beyond keeping a few economists/econometricians in a job!
    5. Brand loyalty of the customers — Multinomial probit
    6. Do people make up their minds before elections — artificial neural network problem ..Basically I look at it as a non-linear regression with fancy functions!
    7. Temperature forecast uncertainty — Use GARCH to judge the correlation structure of the forecast errors

My takeaway : Considering my interest is modeling variables in finance, I think that it would be naive to depend on some linear parameters of an equation with gaussian assumptions in it. Even though a flavor of models are presented in this book, I don’t think models of such type are going to be useful anymore. THE GAME OF GAUSSIAN ERROR TERM is over.