I write at Yahoo: Statistically Losing The World Cup
(Reproduced in Entirety)
“No century-scorer has ever been on the losing side of a World Cup final.”
That was what they said on Twitter, TV and on the internet when Mahela Jayawardene scored his hundred, as Sri Lanka went on to make 274. It was a crippling statistic — the number of times it was repeated almost made me switch off the TV; what was the point of watching the rest of the match if we had lost to the numbers?
If you went just by statistics, there were more reasons for India to lose, just after the Sri Lankan total was posted:
But since India won, we had “beaten the odds”. Did we really? Consider that there have just been nine World Cups in the past. That’s already too little data. Just nine World Cup finals is not enough to say the tenth has “odds stacked against it”. There were no odds to begin with.
Further, out of the nine past World Cup finals, only five have seen centuries. In 1983, when we beat West Indies, no one managed to even reach 50. That means there were only five real data points. Commentators might be forgiven for throwing a sound bite ever so often, because when you hold a mike, you have to say something, but listeners need to separate the wheat from the chaff.
Next, the idea that the past would determine the future, even when the past is based on such shaky foundations, is strange. Especially when it made no sense whatsoever; there is no reason that a team with a century-scorer should always win, and while I haven’t done the analysis, I would be very surprised if in ALL one-day matches, century scorers were always or even more than 80% of the time on the winning side.
Why do I harp about this? Because it happens all the time with investing and trading. There was a scare in August 2010 about a “Hindenburg Omen”, an obscure indicator that has supposedly preceded every major US Stock market crash since 1987. But it hasn’t quite indicated a crash every time it occurs — just 25% of Hindenburg Omen occurrences were observed to have preceded a major market crash. And then, the indication required things like “The daily number of NYSE new 52-week highs and the daily number of new 52-week lows must both be greater than 2.2% of the total NYSE issues traded that day.
The 2.2% always gets me suspicious — is there a reason for the 2.2%, or is it a number manufactured to make the indicator work? Put another way, are we assuming the conclusion and retro-fitting the data on to it? If 2% didn’t work, let’s try 2.1%. No? 2.2%. There. Or this could go on, like, “The number of 52-week highs should be less than the number of sunspots recorded, unless there was a suicide in Manhattan.” Eventually, you will have an indicator, and a brain full of jelly, but not much else.
Many so-called indicators for stocks and indexes take on complex hues, such as taking on moving averages of moving averages and so on. The moving average is simply a “smoothing” function — it gets rid of periodic volatility to tell you the recent trend. But smoothing has its disadvantages; it reacts slowly to sudden changes, so it will only tell you the trend has changed after the trend has changed, sometimes too late to actually take action. A moving-average-based indicator will always be a little late, and you should naturally be suspicious of any ‘formula’ that can predict the next move, based purely on moving averages of price. At best, they can tell you a trend, and if the hypothesis is that the trend will sustain, and that bears out historically in enough instances, you might have a hope with it.
But you can always find a moving average that has predicted the market, using the right numbers and eliminating some inconvenient data by ignoring it — does that mean you’ve found the holy grail? I wish the answer was yes, because I have a whole heap of such formulas invented over the years that are as profitable as used toothbrushes.
“The real estate market has never gone down in any meaningful way” — this statement was often quoted by real estate agents and brokers in the US, and it might have even been statistically valid, with over 50 years of data supporting it. But wasn’t that just correlation? Housing bubbles have been known to go bust in the past, and in different countries. From Sweden to the UK to Greece to even the US in the early part of the century, housing prices have fallen. While the argument is moot today (US House Prices are STILL falling, after more than three years of a downward trend) it remains alive in pockets of the world. Especially the pocket where I live, in Gurgaon, where you can’t lose money investing in real estate because no one ever has.
Eventually, statistics can influence behaviour. If the cricket team believed the past statistic was going to be held true, then they could give up mentally and make it true. If certain technical indicators are believed to work, they will work even more because people buy or sell just about when the indicator says so, marking peaks and bottoms well enough for the indicator to reinforce its usefulness. I often wonder if I look at technicals because I believe in the concept or simply because the trading crowd tends to; the answer is irrelevant.
At some point things break away from the past. Even if you found a statistical indicator that worked phenomenally, could you trust it enough to put all your money into its predictions? The answer, after all the “black swan” events that seem to have swamped us in the last few years, is an emphatic no. As is the answer to the question, “Would you bet all your money on a team led by Dhoni in a World Cup final”?
More Yahoo Columns: