Note: this is a really esoteric bit, but it's the kind of thing I would like to write in the course of my work, if only my editors would allow it. The nice thing about E2 is that I can rock the boat in my mind and suffer no negative professional consequences.
A risk statistic (technically a coefficient) often used in investment analysis, both by professionals (who often use it in conjunction with the captial asset pricing model or CAPM) and individual investors (who think they know what to make of it, but who are usually wrong).
Beta compares two data series--one for an underlying baseline index (to represent "the market") and one for the security (such as a stock or stock mutual fund)--to determine how closely their individual swings are related. Mathematically speaking, it's the covariance divided by the variance. The baseline (for US equities, the index most often used for this purpose is the Standard & Poor's 500 Index) has an assumed beta of 1.00. Customarily, what is compared are monthly returns for the index and the security over a three- or five-year period.
The generalization for beta is that it shows how a security performed relative to the market (i.e., the baseline). A stock with a beta of 1.20, for example, is said to--in general--have risen 20% higher when the market was up and fallen 20% farther when the market was down. So if the market was up 10% for a certain year, you would assume that the stock with the 1.20 beta would have been up by 12%. It is also generally claimed that a 1.20 beta shows that a security was 20% "riskier" than the market, assuming a 1:1 ratio of return to risk.
Unfortunately, the generalization is crap; and as a professional financial writer, I detest it. Beta is specifically designed to show the level of systematic risk displayed by a security: that is, how much of the security's price fluctuations may be attributable to general forces affecting the entire market. Thus, a reasonable way of looking at the stock with the 1.20 beta is to say that it is 20% more sensitive to broad market effects than the market itself. So if a panic hits Wall Street, it's safe to assume that this security would take a price hit 20% greater than that suffered by the Standard & Poor's 500 Index.
Or is it? The implicit assumption here (and one made by legions of investors) is that every security in the marketplace is exposed to the same sort of influences. This assertion does not stand up to serious scrutiny. Small stocks tend to respond in entirely different ways to certain stimuli than the stocks of large companies. And these days, it seems like whatever's good for tech stocks is bad for the rest of the market.
In truth, the past movements of securities may or may not be closely related to those of the broad market. A great indicator to use is R-squared, which examines the correlation between two data series--in this case, between a security and its baseline. If a security has an R-squared near 1.00, it means that the security does have a close relationship to its baseline index. As R-squared figures decline from 1.00, the relationship becomes less clear. In general parlance, it is fair to say that a stock with an R-squared of 0.85 "shares" 85% of its fluctuations with the baseline. This is a reasonably close relationship, but a security with an R-squared of 0.25, for example, actually has little in common with the baseline.
It is possible, and I have tried to find the "ideal," to construct a data series that has a beta of 1.00, but an absolute risk (measured by standard deviation) that is off the charts compared to the baseline; and, moreover, has an R-squared approaching zero. That is to say that it's not difficult to show that beta can be misleading in the extreme if it is examined in a vacuum. Investors who use this figure would be wise not only to familiarize themselves with its statistical underpinnings, but also with those of R-squared.
Remember, profits in the stock market come only by making profitable use of the information at your disposal.
Sockpuppet has quite rightly pointed out that beta is often used at the portfolio level by money managers who use it as a rough benchmark for their performance/risk vs. their peer group. But, in these cases, you'll often find them computing a beta vs. the Standard & Poor's 500 Index as well as against an index that more closely approximates their market--for example, a small cap fund manager may compute his fund's beta using the Russell 2000 Index as a baseline. This is an extremely appropriate use of beta.