Mean Reversion and Time Series Equations

This page explains time series equations that underlie Monte Carlo simulation and in particular the underlying formulas for volatility, mean reversion and other factors. Time series equations must have some kind of stochastic element, but they can include a host of other variables, the most important of which is mean reversion.  Factors other than mean reversion can include minimum and maximum conditions that reflect periods of surplus capacity.  A basic time series equation can be expanded to include sudden jumps, mean reversion, minimum levels, maximum levels and trends.

The notion of making equations could have deep philosophic elements. In your life there will be random events such as pandemics, economic recessions, medical emergencies, dog bites, political upheaval … Your task in life is to manage these uncertain events and have a reasonably satisfying life. You can try to get some mean reversion with an education, an open mind, different debt levels and other equations on top of the random events. Maybe you should think about Monte Carlo and time series equations as more of a philosophy than some kind of practical statistical analysis that you will be able to use.

Volatility and Standard Deviation of Rate of Return

Mean reversion is a crucial factor in developing time series. Efficient markets theory suggests the stock prices do not have mean reversion because all of the relevant information is supposed to be incorporated in the current price. This suggests that if prices for example change according to a moving average, you could beat the market by simply buying when the stocks go down and then sell when stock go up. If you could do this and ignore the real information that is changing the value of stocks, the market would not be efficient.

Efficient markets without mean reversion can be modelled by computing the standard deviation of the rate of return. Remember that growth and and rate of return are similar concepts and the standard deviation of the rate of growth is essentially the same as the standard deviation of the rate of return. I wish I would have learnt about volatility by just understanding that volatility is the standard deviation of returns. The idea of a non-mean reverting series representing an efficient is that you look at the standard deviation of returns rather than the standard deviation of absolute returns. The efficient market returns can be represented by a discrete equation or a continual equation just like any growth or return analysis.

Other time series ranging from the weather to oil prices to interest rates are presumed to have mean reversion. For example, if it is very hot one day and the average for prior years is less, then you can predict that the next day or the next week will not be so hot. For oil prices, there are various changes in demand, supply, political events and other market forces that can push prices up or down. But if prices move below the marginal cost of production, oil producers will stop production and the prices will have a floor. On the other hand if prices are really high, then more production will come into the market and the price increase will be limited. The mean reversion can be modelled with the following formulas:

Price (t) = Price (t-1) x (1 + g)

Price (t) = Price(t-1) x EXP(g)

I should be able to tell you that EXP(.05) = 1.05121. This is the growth rate over a period in really tiny increments where you keep on compounding. In the discrete formula, all of the growth occurs at the end of the year (this is like using the opening balance to compute interest in a debt balance calculation. To see how the normal distribution works with probability, I have made a very simple example where you can plug in the expected return and the standard deviation of return which is the volatility. The range in a normal distribution is about +- 4 standard deviations from the mean. So I make a simple list using EIS and you from -4 to +4 with an increment of .25. Using this you can multiply the standard deviations from the mean by the standard deviation and add the expected return. Then, you can see the distribution of potential returns as a function of the volatility.

When you think of standard deviation you can try to remember the ideas of a normal distribution that only depend on the mean and the standard deviation. I can never really remember if 68% of the predicted values fall between +1 and -1 of the standard deviation. So you can quickly check this in excel.

Price(t) =

Price(t-1) + A x (Mean – Price(t-1) ) +

Volatility x (Price(t-1) x Normsinv(Rand())

When applying this formula, A is the mean reversion factor and it is essential to put the mean price in the equation first and not last. For example, assume that the mean price is 100 and the current price is 150 and A is 20%. In this case you want the next price to go down and not up. So it is important to put the 100 first in the equation which yields 20% x (100-150) or the price moves by 10 back towards the mean. If you are modelling weather with 100% mean reversion from one year to the next, you could put in a mean reversion factor of 100%, meaning the factor would be 100%* (100-150) or -50. In this case it would be better to measure the volatility as the standard deviation divided by the average.

The video below explains how to incorporate various factors of time series equations into an excel model.