Any experience with Forecast Pro/Box-Jenkin forecasts?

[Dan Leontieff]

Hi

Has anyone ever used the ARIMA (Box-Jenkin) model to forecast time series, either price or volatility, e.g. ^VIX ?

From what I've read, it's supposed to be a pretty good approximation of short-term moves. It's forecast horizon is 14 days.

It's automated in Forecast Pro, so you just push the buttons, without doing the actual math which is pretty involved.

Anyone has experience with Forecast Pro and/or using autoregression models (ARIMA, GARCH) for time series forecasts?

Thanks

Dan

[astra]

Hi

Has anyone ever used the ARIMA (Box-Jenkin) model to forecast time series, either price or volatility, e.g. ^VIX ?

From what I've read, it's supposed to be a pretty good approximation of short-term moves. It's forecast horizon is 14 days.

It's automated in Forecast Pro, so you just push the buttons, without doing the actual math which is pretty involved.

Anyone has experience with Forecast Pro and/or using autoregression models (ARIMA, GARCH) for time series forecasts?

Thanks

Dan

Hi Dan,

Your name rings bells from the past for me! Well, I have it written in my to-do list for some time now, but always something else pops up... so keep in touch...

I think, the issue with this type of research is not whether an X software makes the task easy or not, but what inputs you will feed to the model(s), and this is not an easy task. An understanding of the aims (not necessarily the maths) of the models is required and this can take time.

Regarding the software, I think you can find very dissent open source software nowadays to do this type of work, so why to pay for a commercial one? - google the terms and see what comes up.

[Dan Leontieff]

Thanks, excellent news.

Leaving the philosophical side of it for now, any idea if it actually works?

1. Can it give a sense of direction of the price series, up/down or sideways?

2. Can i use the confidence intervals, say, to trade options?

thnx

[astra]

Thanks, excellent news.

Leaving the philosophical side of it for now, any idea if it actually works?

1. Can it give a sense of direction of the price series, up/down or sideways?

2. Can i use the confidence intervals, say, to trade options?

thnx

Hi,
To the 1st question I answered it before. It depends of the inputs you feed to the models. I recommend you to check http://www.wilmott.com/ - a top forum and resource site for quants.
2. If you are trading options, NT is not the right place for you. Unfortunately, you can't trade options with NinjaTrader:

http://www.ninjatrader-support2.com/vb/showthread.php?t=14687&highlight=options+trading

Hope it helps,
astra

[sefstrat]

Does it work? yes.. but as Astra mentions these models are only useful if you understand the implications of the model.

For instance, if you fully understand what GARCH implies about volatility clustering across different time intervals, then you can create a very powerful system without any complex math (assuming you have the patience to code multi-timeframe strategy in NT ).

If you do understand the math (or want to learn) there are much better function approximation methods now than box-jenkins approach, for instance recurrent neural network based or gaussian process regression.

[Dan Leontieff]

Astra, thanks for pointing me in the right direction.

My math skills are pretty basic, but a lot of trustworthy people think these autoregression models have value, so I am trying to check these out from a trading perspective

Will check the Willmott forum though, I am curious what mathematical finance people are saying about this. Thx

Dan

[Dan Leontieff]

Sefstrat, thanks for your thoughts on this.

I don't have the mathematical firepower to deal with the mechanics of the models themselves.

All I know is that GARCH is used for volatility forecasting, that volatility is easier to forecast over longer periods of time, and that the model is based on an interplay of three forces: 1 mean reversion 2 serial correlation 3 momentum.

In terms of the Box-Jenkin, I am just trying to figure out if it is any better, in practical terms, than other things.

For example: http://www.hoadley.net/options/barrierprobs.aspx?
This assumes a directionless market, random walk etc., but it's a surprisingly effective approximation of actual price movement, or at least range, over a certain period of time (assuming you know volatility and it's constant).

[sefstrat]

Np, looking back at your previous questions I might add that ARCH models are not really about forecasting the direction of a move but rather the extent.

You may want to check out the olsen & associates classic intro to high frequency finance.. lots of good info there on these topics

http://www.amazon.com/Introduction-High-Frequency-Finance-Michel-Dacorogna/dp/0122796713

[Dan Leontieff]

Super, thanks for the info guys!