Top critical review
6 people found this helpful
The 20-year disadvantage
on May 25, 2013
Caution - this book will leave you less informed about predictive analytics than you likely were before. So pick up at your own risk. Several issues.
1) The sensationalist title and description suggest we'll be reading a useful treatment of the role predictive analytics plays in today's economy (and, that there's something particularly NEW about predictive analytics, like some recent breakthrough; or some heretofore unknown magic to the "2 seconds" referenced in the title). What we actually get is an extended blog post of randomly mixed tidbits and stories of anything where someone used a computer or a calculator to take a guess on the future, without much of an overarching thesis or structure. And, worse, mostly just stories from 10 years ago. For example, we have a guy from Reliance talk proudly about how his crack data analysts have created a churn model that predicts which mobile phone customers are likely to churn. If you think that some low-factor logistic regression model is somehow a newsworthy predictive analytics breakthrough, you've slept through your first 15 minutes of computer science classes in community college.
2) The churn prediction story at least is so simple that the book can't mess it up by its usual approach of uninformed ultra-high level discussion. Unfortunately, that's what's happening with most other stories in the book: an amateur's look at a complex topic, plus some hollow cheerleading about how more data would be cool in that situation. For example, one chapter "discusses" (not sure if I'd call it a discussion) how Wall Street traders now have models that predict what will happen in the markets. Aha. The next chapter then suggests that the Fed messed up the financial crisis of 2007 because it only finds out every 6 weeks how the economy is doing, by asking a bunch of local businessmen "how's business". (Presumably it is referring to the Fed's Beige Book.) Finally, how cool would it be if the Fed had a real-time monitor of the economy, like those Wall Street traders' models? Now - that is about the dumbest thing that I've read about the financial crisis. Take it from a former hedge fund guy. First, the Fed (like the rest of the market) obviously has a lot more frequent economic indicators than the Beige Book. Unemployment claims are released weekly; the financial crisis was presaged by a drop in home prices accurately reflected in the monthly Case Shiller index (which was also traded in the futures market); etc. Plus, the Fed has its fancy DSGE model of the US economy which surely beats the dumb Reliance churn indicator celebrated by the book in complexity by orders of magnitude. And finally, Bernanke can just pull up the stock and bond markets on his Bloomberg (which he does all the time), to see how the markets think about how the economy is doing. Second, to suggest that the Fed should just have used some of those predictive models used by traders, and it would have prevented the crisis, is just hllariously dumb. The quant hedge funds were the first ones to blow up (in August 2007!); and it was precisely forward-looking trading models that drove a significant part of the crisis. (Plus, the few hedge funds out there that did have smart views of what's happening with money and credit once the crisis played out were actively consulted by the Fed all the time.) That's not to say there isn't a hugely interesting discussion to have had about the value of economic models, whether the financial markets are really good at discounting future economic conditions in today's prices, and whether there'd be value to more real-time data sampling across the economy. But this book isn't adding anything to that discussion. It takes more than 2 seconds to think about these topics.
3) Finally, much of the book is made up of quotes like "as Wired wrote in 2011...", "as Ray Kurzweil wrote...". That's not a book, that's a blog post. If you already don't have a particularly new thesis to offer, at least compile a book that's more than just about stuff you've read elsewhere or heard someone say.
Big data IS a big deal, the abundance of data we have today in more and more industries is a relatively new thing, and our ability to manage vast quantities of data through column-oriented databases or map/reduce systems has indeed gone up. So there is a lot of room for smart discussion. That's not taking place in this book. If you want to read about the limits of statistical models, go and read Nate Silver's recent book. If you want to read about predictive models, pick up an introductory computer science book. Those at least give you information on where we stand today, and not 20 years ago.