Is Now the Time to Buy Stocks?
online.wsj.com
online.wsj.com
I did an analysis of this recently, available at: http://saffell.wordpress.com/
My conclusions:
1. Now is a better time (on average) to make a 5-10 year investment than a random point in time is.
2. For 20-30 year investments, the recent fall in prices make little difference to returns over such a long period.
This is all based on historic data (from 1871 to 2008), which, as we all know, is no guarantee of future performance..
Some may consider it linkjacking, especially since I posted this once before on HN (a few weeks ago) when i first wrote it
The work is all original.
PS: I keep wondering what all those baby boomers taking money out of 401k's to retire with is going to do to the market.
Dividends are included by assuming that all dividends are reinvested w/o taxation. More details on the blog.
* First, your "N" of 1,578 is way overstated, to the extent that we should be inferring that we're looking at 1,578 independent data points. What you seem to have done is to treat every month from January 1871 through June 2002 as a separate data point. However, the past 12 month performance for any two data points within 11 months of each other are correlated (highly correlated the closer in time they are). Further, the 5 year returns of any two data points within 59 months of each other are also correlated (again, very correlated for close-together data points). So you actually have just 26 independent data points here, which is a statistically dubious sample size. For the 10, 20, and 30 year versions, you have 12, 5, and 3 independent data points, respectively -- statistically meaningless.
* Second (and for the same reason), your N for each subset of the data corresponding to different timing strategies is even less. Without looking at the historical data, I can't say exactly how many independent data points you really have in each subset, but I know, for example, that in the last 137 years there have been only a few stock market declines of a magnitude of 20 - 50%; certainly nowhere near the 139 that your "N" suggests. So even the 5-year study has a meaningless sample size for the individual triggers.
* Third, your subset samples are biased because individual "real" data points are reflected in varying quantities depending on how many adjacent months the signal remained for. So, even if we had enough real data points in theory, your sample is skewed towards those buy signals which lasted longer.
* Fourth: you say, "For 10-year investments, investing when the market is down 20% to 50% is probably going to produce better returns than a investment at a random point in time, but it's a close call". But even if your data were accurate, the 10-year charts are such a close call that there's actually no conclusion to be drawn from them.
* Finally, you had no prior hypothesis. This is a fatal flaw with many attempts to mine historical stock market return data. Instead, you measured sixteen separate investment strategies (four different buy signals times four different time horizons). Without a prior hypothesis, you have not done a scientific study, but you have just mined data for patterns. In any sufficiently large corpus of data, there will be patterns simply by coincidence. A prior hypothesis is necessary to distinguish those patterns which exist by coincidence from those which we predicted ahead of time due to a correct explanatory framework.
Even if your analysis had none of these flaws, we'd need to be wary of it, because of the phenomenon of publication bias (only those studies which show something interesting get noticed by people; but every so often someone predicts something, and the data seems to confirm it, simply because they had an unusual sample despite their having done everything properly). So your conclusions would merely be tentative pending followup studies (on independent data points; which means we'd have to wait decades to start to be confident). But each of these flaws, individually, are enough to completely invalidate any inferences we can make from your data.
Don't feel bad, however. People have been drawing un-supported inferences from stock market data forever.
Now that we have that subtlety out the way, lets look at the more substantial issues. Let me take these one at a time to explain my thoughts:
1. Yes, you're right, there is a lot of correlation. While there are 1578 months, there are far far few economic cycles since 1871, which is another way to think about the problem you are exposing (also in 2). However, this is the data that we have, and I am taking the approach of 'something is better than nothing'. I would say that what this analysis does do is to answer the question "what would the distribution of returns look like for an investment of x year, using timing strategy y, if employed since 1871 to 2008?". Would you say that is fair? If so, the next question is, "can this be generalized in a way that makes it a useful tool for predicting the future?" My answer would be, "not with significant confidence", which I think is your point - is that right? The next question would then be, "so why bother doing it at all?", and my answer would be, "intellectual curiosity" and maybe, "because it's better than nothing at all", though I can see why you would dispute the latter.
2. This essentially is the same as (1), so again I agree with your observation, but maybe not your conclusion. The "n=139" means that in my sample there were 139 months between 1871 and 2008 where the monthly price was 20%-50% lower than max price from the previous 12-months. I'm fairly sure they're accurate.
3. I'm not sure I understand - please explain further.
4. I think we totally agree here on the statistical validity of the data. Where we differ slightly is the presentation. I have not made any conclusions (again my comment on this page was misleading, but at least provocative :). This is a blog that I am using to get other people to help me answer my questions. It is not a research paper. I talk about what I see, not what I conclude. I looked at what my conclusions would be because I know that when most people read this the easiest way for them to grasp the results is to see a viewpoint of what those results would look like if translated to conclusions. I.e. the "would conclude" comments are a tool to spark debate, which seems to have worked:) That is an easy thing for me to say, I know - it may even sound like a cop-out, but if you critique this post as you would a research paper then you miss the point.
5. I could not agree more. In fact, I make this point often to others. I actually did have a hypothesis, but it wasn't crisply defined, nor did I write it down, nor did I mention it in my blog. All of this is bad practice, I know. I didn't go into this on the post because I wanted to make it more accessible (and shorter), by quickly getting to the results. Again, that's a style choice, and I would never dream of doing it in a research paper.
6. (on publication bias). Again, I wholeheartedly agree. Independent verification is a corner stone of good research - but someone has to go first.. If anyone does ever choose to followup on this then I shall be deeply flattered, and eager to see what comes of it. Thank goodness I'm not making any conclusions.. :)
7. I don't feel bad. I'm grateful that someone has taken time to give detailed feedback.
You drew conclusions from the data. It's not advice, but the conclusions were unsupported, so I consider it unsupported data mining. I'm not questioning your motives. :)
1. Yes, the appropriate question is whether we can generalize from this data. One should prefer uncorrelated data points when sampling data, which is why I say you only have 26 data points. Is it "better than nothing"? Depends on what conclusions you draw. "Tea leaves" are better than nothing if you just think they're pretty. But if you think they have meaning, then they're worse than nothing. In this case, the study has zero predictive value because there is no scientific result.
2. I understand what the N means, but it's misleading since in statistics one usually thinks of N as referring to the number of independent data points.
3. What I mean is: a "real" data point would be an actual market decline of 20-50%. In your model, some of those declines may be represented only once (say, because there was a market recovery the following month), and others are represented up to 11 times (because there was no recovery for the next 11 months). So not only do you have a (much) smaller sample (by a factor of between 60 and 360 depending on which analysis) than your "N" reports, but you also have a biased sample because each buy signal is represented for a varying amount of time depending on how many months it lasted for. This means that, of the handful of market declines which actually occurred in the last century, you might really be representing an even smaller sample.
4. You did draw the following conclusion: "For 10-year investments, investing when the market is down 20% to 50% is probably going to produce better returns than a investment at a random point in time, but it's a close call".
Anyways, thanks for your reply. Sorry if my initial comment was a bit too snide. :)
In a market where all stock prices have fallen dramatically, you'll be able to buy it cheaper than before - but note, it might still be too expensive.
The trick is in recognizing a good business and valuing it - but that's much more doable than deciding whether to "buy stocks", because it's more definite and specific.
To put it in perspective: it's what ycombinator does.
For normal investors, in ordinary times one should pick up an index fund (roughly one third of your total investments, the rest of them are not stocks) and hold unto them until you retire. The formula can be bend slightly depending how far you are from retirement.
Of course these are not ordinary times.
But enough to buy a controlling stake in mid sized companies, that not.
Happy middle's where you want to be.
With a million or less, he could assess the small companies that aren't covered by analysts and so on, to find one that was (1) great (2) cheap
The lack of money is not the deficit that stops you (and me). Mostly, it's effort, and self-confidence in your own considered opinion when it is contrary to everybody else. To buy when others sell - because you've done the research, and trust your judgment. It's mostly an emotional or moral quality. If your judgment is wrong, you have a lovely learning experience that cannot be obtained in any other way (and Warren Buffett has had that experience).
Of course if he announced which stocks he picked at the time he purchases them the price would probably increase. Heisenberg and all that...
I often wonder if people like Jim Cramer end up manipulating the market just by talking publicly. And is it illegal? If Cramer says "I just bought such and such stock and think it's a great buy", and the price goes up because of it, and he sells his shares for a profit, is that illegal? He simply stated a fact and his opinion. Where is the line drawn?
The Berkshire Hathaway stockholders might object. It's a dumb objection, but it's probably enough to get past summary judgement.
There are examples of even great and old big names falling to ashes right now: Lehman Brothers, AIG, and GM.
http://en.wikipedia.org/wiki/Dow_Jones_Industrial_Average#Fo...
And you lose a lot of mobility on most index funds, I'm still fighting Fidelity to get my money out of one! Don't make my mistake, Dilbert should stick to cartoons.
They either didn't base their argument well or are completely delusional and can't read their own graphs.
And this current crisis is all about prediction based on past data can be plain wrong. See: http://news.ycombinator.com/item?id=360412 The End of Wall Street (HN main page today. Beware it is very long!)
Unless you do it algorithmically, it's incredibly boring, but once you have started it's hard to stop. Market timing is an addiction :-|
appreciate your downside risk against your upside potential. if you fear the market will go to zero than we will all have much bigger problems than counting our returns.
if you have any faith in the american model than you must conclude that this too shall pass.
PS: Options on GM stock could be a good investment of you are willing to take a lot of risk.
the point is that things are relatively cheap where we stand today. they may get cheaper, but they may not. look at your downside risk (missing 25% of a bigger bottom) vs. upside risk (we've already hit bottom).
or, if your nicholas nassim taleb, you think that you'd be crazy to make any decision based on the graph, because tomorrow everything could go to zero, and you'd be broke.
i'm thinking of shifting part of my portfolio to stocks. but given volatility, like the article says, it's going to be a small segment, and i can allocate more as it goes up. but i'm investing in the long term, and if the S&P goes to 0 then i won't be worrying about money. i'll probably be worrying about people trying to break into my house and steal my cans of campbell's.
Buy the fear, sell the greed!
The other explanation is that investment firm XYZ is wrong when they say that a particular asset is undervalued.
I disagree on two counts:
1. The market has seen no significant net increase in price, and is down significantly on recent (12 month) highs. So one may be early for this cycle (or at bottom). Alternatively, one may be late in the sense that there will be no further investment opportunities ever in the future, and that one would have needed to invest during a previous cycle:)
(This is see as evidence that it is not too late)
2. Markets act on fear and greed. As Buffett said, be "be greedy when others are fearful". Sentiment changes slowly, and heterogeneously (by which I mean different investors change their sentiment at different times and speeds). My observation is that most people are still fearful, but some are not, and the author of this article (and Buffett) are two such examples. So the logic "one person calls 'buy' -> no opportunity" does not hold for me. Now, if many (>=50%) are calling 'buy' then it is time to be fearful.. (This I see as rationale for why your statement isn't true, but not evidence)
Both are cheap, both will double in a year.
Mark my words.