Forecasting Algorithm

toddyl

Hi,

I am developing an internal application to forecast CPU usage over time. As this fluctuates similar to Equities I thought that I would post in this forum.

Basically what I have is an average and max value for each day for the last month. From this data I would like to apply a forecasting algorithm which would be able to predict from the supplied info the possible behaviour in future.

Are there any common economic forecasting algorithms that could be used in this scenario?

Thanks for your help,

Tom

You know that economic forecasts are theory based right?

gerry87

Moves like equities how? Does it follow a completely random walk? What determines cpu usage? Could you model that instead? Has usage been steadily growing over time like equities? Do you have any graphs of the data you could post?

Forecasting random walks with any accuracy is pretty much impossible else it wouldn't be random! A simple but inaccurate way would be to just average out the change from the first observation to the last one in your data, set that at a future daily growth rate, if you have enough data it could work.

It's kind of unclear what you're looking for. What do you need it for?

You may be better off over in the maths forum asking about stochastic processes and/or monte carlo simulation.

10000maniacs

Go to the Start button in Vista and type Perfmon in the Search bar.
However, how you can say that CPU usage fluctuates in a similar manner to equities is beyond me.
By the way when you mention an acronym like CPU, put its meaning in brackets. You are just being ignorant otherwise.

leninbenjamin

10000maniacs wrote: »

By the way when you mention an acronym like CPU, put its meaning in brackets. You are just being ignorant otherwise.

in fairness, this is the internet. anyone who doesn't know the meaning of that acronym shouldn't really be here.


:P


OP, you do realise as well that this recession has confirmed that most forecasting and charting for equities is absolute horsh*te?

silverharp

leninbenjamin wrote: »

OP, you do realise as well that this recession has confirmed that most forecasting and charting for equities is absolute horsh*te?

That is a generalistion, for instance Dow theory gave a sell signal in 07. Otherwise looking at charts, the US housing stocks were the first to turn, then subprime lenders, subprime debt spreads increased, Then the larger banking stocks started to turn, Banking stocks lead the rest of the market so a crash of other stocks was pretty much set in stone.
The above isnt hindsight on my part, I was looking at this stuff as part of an unfolding process and many investors / analysts were looking at the same thing. The only problem is that the market is dominated by "dumb" long only pension funds

As for the OP question it is not possible to chart the next days direction on a reliable basis, however it is possible in terms of stock market charts to find repeating patterns that have predictive value however obviously not 100%

leninbenjamin

silverharp wrote: »

That is a generalistion, for instance Dow theory gave a sell signal in 07.

Dow theory? was that not proven to be a pile of crap or am i wrong...

and fair enough the markets were strung along by the banks, but that doesn't make a good forecasting algorithm...

silverharp

leninbenjamin wrote: »

Dow theory? was that not proven to be a pile of crap or am i wrong...

I havnt really followed it in detail to be honest, I've treated as as aside "weather forecast" Tim Wood is one of the main analysts that uses it.

There are many ways to look at chart intenals that can be indicative of upcoming trend changes, you can use Algotithms to look for markers but no it is not possible to run charts forward


http://www.safehaven.com/article-9031.htm

December 15, 2007

Dow Theory Update
by Tim Wood



The primary bear market confirmation that occurred on November 21st when the Industrials confirmed the Transport's break below the August closing low remains in place in spite of the recent rally. In fact, the rally that began out of the November 26th low was anticipated and I stated before this rally began that it was going to cause many to question the integrity surrounding the November 21st Dow Theory primary bear market confirmation. That has definitely proven correct.

Robert Rhea, who was the leading Dow theorist in the late 1920's and 1930's stated: "Charles H. Dow never intended his theory of price movements to be construed as being a method whereby the royal road to riches could be found. While the principles laid down by him and developed by William Peter Hamilton may assist us in understanding something of security price trends, it would nevertheless be a fallacy to undertake any discussion of the subject without making the point very clear that no dependable method of beating the market has yet been discovered. Intelligent observation and study of the ever-recurring formations in the averages which Dow and Hamilton noted can, however, prove invaluable to both investors and speculators."

In reading the volumes of original writings on Dow theory I totally agree. Though not infallible, Dow's read of the averages were meant to be a "Barometer" for gauging future finance and business conditions and one of the very basic concepts of Dow's theory is that the averages look ahead and discounted everything. Hamilton wrote "The fluctuations of the daily closing prices of the Dow Jones rail and industrial averages afford a composite index of all the hopes, disappointments, and knowledge of everyone who knows anything of financial matters, and for that reason the effects of coming events (excluding acts of God) are always properly discounted in their movement."

As an example of this, the non-confirmation that was formed between the averages in October and that can be seen in the chart below warned that something was wrong in accordance to Dow theory. That non-confirmation then evolved into a full-blown primary trend change on November 21, 2007. As I read the averages the last primary bearish trend change in which the averages forecasted stormy conditions occurred on February 25, 2000 and following that confirmed primary trend change the Industrials ultimately fell by over 27% from that date and by over 37% from their January 2000 closing high. According to Richard Russell's read of the Dow theory at that time, the primary trend change occurred on September 23, 1999, which was prior to a final unconfirmed push to new highs by the Industrials. In either case, regardless of whether the last bearish primary trend change occurred on September 23, 1999 or on February 25, 2000, the outcome ultimately ended very badly.

Mikel

Dow Theory? LOL!
OP, if you mean you want to model CPU usage as a stochastic process, using an equity model would not be suitable because equities are modelled with an upward drift.
Maybe you would look at a mean reverting (OU) process, you could incorporate sudden jumps using a jump diffusion process (Variance Gamma), tbh the maths forum would be the best place to ask

EDIT: if you wanted to model it simplistically, you could use a brownian motion, calculate the 'volatility' from your historic data, maybe use a lognormal distribution so your usage can't go below zero and run a monte carlo

babytooth

take a semi stochastic process with a decreasing coefficent on the further time lag coeffs in your eqtn, you would also need to have a coeff of vol for historic vols that also has a mean reversion process....

Not an easy one to create....