Market Risk Premium

The market risk premium is a financial concept that is generalized as something like this:

A broad portfolio of risky assets, so broad such that risk from any particular asset can't really have any material effect, returns in excess of the "risk free rate", which is what you earn on securities with "no risk". This risk premium compensates you for providing capital to projects with a risk greater than "no risk".

That's pretty hacked up, but its the jist of it. Most people agree with the above. The "risk free rate" is usually stable government short term securities. In my humble opinion, these aren't risk free, as they don't seem to price in the possibility of government collapse from aliens attacking us (see how close they came to winning in Independence Day?). Or maybe they do price in this risk, and I just haven't been able to empirically measure it. I'll leave that for another day.

Now measuring the risk premium is different. There's lots of empirical data, but really it's all data mining. What really is a broad portfolio? Over what time period? There's lots of defendable answers to both, so there's lots of defendable answers to the risk premium question.

The bottom line is a model is guesswork. It's a simplification of reality by definition. It's an attempt for mankind to lay a framework of orderly thought over things that may or may not be orderly. Good models generate plausible forecasts. Forecasts are never really right anyway...

Measurements of the risk premium have been done in numerous academic settings. There's support for a range of values, but usually I see between 6% and 11%. That means, on average, over time, broad portfolios of risky assets (ie. this is usually assumed (yes, another assumption) to be something like the S&P 500 or Wilshire 5000 stock indexes [side note: do broad stock indicies represent diversified asset classes?]) return either 6% above the risk free rate, 11% above the risk free rate, or any number in between those two, or if you pick a peculiar time period and do cruddy statistical work, any number outside that range as well.

And why does anyone need such a concept? To look forward to the future. To make estimates and educated decisions. But when was the last time history predicted the future accurately? Ok, enough on that mental hurdle for now.

What's the point of this? Pick a number between 6% and 11% and stick it in your models. Then show sensitivies run at other percentages. Why? Because you aren't going to magically find the "right" number after midnight with your investment banking co-workers. There isn't a "right" number, just like there isn't an exact, agreed upon average temperature for the month of April. Daytime? Night? Highs? With or without windchill? What time period? Where?

It's like arguing with your spouse, it's an unproductive sink hole of time.

I'm off to navel gaze on the average temperature of April.