Having a case of recency?

[larryswedroe]

http://www.cbsnews.com/8301-505123_162- ... their-gut/

the piece discusses the all too human tendency to put too much emphasis on recent/short term results and ignore long term evidence

Best wishes
Larry

[Johm221122]

I would be the first to admit it is hard to not think about recent investment history, I try to just turn off the noise
John

[nisiprius]

Interesting blog posting you linked by Jared Kizer on The Small Sample Problem. I think this is a huge, huge problem for investors and investing theory. I have gone through these stages in my personal progress toward nihilism.

Stage 1: Wow, "everybody's" talking about it. The Money magazine in the dentist's office. The glossy thing I get quarterly from the brokerage. The guy at work who subscribes to The Wall Street Journal. Maybe I'd better get some of those (130/30 funds, oil ETFs, international bonds).

Stage 2: I am a disciplined, long-term investor. I shall ignore fads. I shall pay close attention to "track records" and "consistency," as shown in the 10-year charts and numbers in the fund factsheets. Nothing less than ten year for me!

Stage 3: Ten years is nothing. I only believe it if you can show it to me in CRSP-based data going back to 1926. (Thank heaven CRSP made those adjustments to correct problems in their data...)

Stage 4: It's hopeless. By the time there's enough data to believe the parameters can be estimated. the world has changed. Is it really possible to the believe that the stock market of the 1920s, when a comptometer was the latest computing technology and there was no SEC, is really the same thing as it is today? And besides, what if Mandelbrot is right? You should probably have both some stocks and some bonds--nobody actually knows anything more than that.

Many are curious about the future prospects for bonds, but what do you make of something like this, I mean what do you make of it?



So, what's the next one going to look like? We have a grand total of 2-1/2 data points in 140 years. So, I pulled 2-1/2 chips out of the bag, and one of them was "5" and one was "15" and the broken one was "5 point something." 15, 5, 5.5. Put THAT in your T-test and calculate the 5% confidence limits on the mean!

[Call_Me_Op]

Nisi,

I like that long-term yield chart. I agree that it is impossible to use is for any predictive purposes - but I like it anyway.

[whomever]

From the article:

If you're like most people, you would probably guess bag A, since 80 percent of the chips you withdrew were red, versus just 67 percent from bag B. However, you're more likely to be right if you chose bag B. The reason is that because the sample size from bag B is much larger, you have more confidence in the result.

I'm not sure I follow the logic of the statistics here. You have two estimates - for bag A the estimate is 0.8 (with a large confidence interval because of the small sample size), while for bag B the estimate is 0.67 (with smaller CI because of the larger sample). But the best estimates you have are still 0.8 and 0.67; I don't see that the available data supports the idea that the true fraction for B is greater than the true fraction for A.

(Moving from pure probability to applied investing, there is certainly a 'minimize the worst case' argument for choosing investments; if the available data estimates that investment A returns 5% plus/minus 2% (because it has a long track record) it may be a better choice than some brand new investment B for which the limited available data can only estimate as returning 10% plus/minus 10% - but that's a slightly different argument than the quote seems to be making.)

Maybe I'm being a statistical pedant ...

(just to be clear, nice article, and I agree with it's main thrust)

[sls239]

I'm not sure I follow the logic of the statistics here. You have two estimates - for bag A the estimate is 0.8 (with a large confidence interval because of the small sample size), while for bag B the estimate is 0.67 (with smaller CI because of the larger sample). But the best estimates you have are still 0.8 and 0.67; I don't see that the available data supports the idea that the true fraction for B is greater than the true fraction for A.

It is because you are told that one bag is 1/3 red and one bag is 2/3 red. So you aren't trying to make a best guess of the percent of red in each bag, you are trying to guess which bag it is that has 2/3 red.

What you have to compare is the probability of getting the result you did if Bag A's true value is 1/3 and B's true value is 2/3 versus the probability of getting the result you did if Bag A's true value is 2/3 and Bag B's true value is 1/3. You will find that getting the result you did is much more likely if A's true value is 1/3 and B's true value is 2/3.

In other words, statistics aren't useful if you aren't asking the right question anyway.

[whomever]

It is because you are told that one bag is 1/3 red and one bag is 2/3 red.

Ahhhh... didn't see that in the article (my reading was that those were the proportions, not that you knew them).

[sls239]

Yeah, I don't think it is worded quite right because he wanted the reader to know the answer. It becomes clear he is talking about a "which" question when later discussing the asset classes because that is another example of asking "which."