Jun

10

I often think about the concepts in the title of this post. So, while waiting to have my positions decimated by a 'tape-bomb' from the embarrassing, puerile discussions between the Greeks and their creditor protagonists, a brief discussion might be in order.

Shapes: When looking to predict markets using information from other markets, my experience is that it may not be too deleterious to use triangles (in some very complex relationships the square or rectangle may be helpful).

So, for example, I might use a triangular shape with the three corners consisting of two predictors and one market to be predicted. The use of the rectangle or square adds another predictor but I believe in this case that too many cooks may spoil the broth (at least in terms of the out of sample testing before entering production/ live trading).

Intriguingly, the techniques that make money in the triangular framework are decidedly non-linear relationships between the three markets. In this context 'non linear' would mean that the chart of the predictive function would not resemble a straight line but some other look.

Univariate versus multivariate question:

It is my contention that futures markets whose underlying asset is something such as stocks, gold, oil, bonds (i.e not currencies) have characteristics such that prediction is best done by looking for multivariate relationships. For example, you might try to predict stocks using bonds and oil rather than just using past SP500 futures information to predict the future SP 500 futures movement.

Of note, and perhaps why currencies are considered so difficult, is that multivariate prediction in currencies, in my experience, comes a very distant second to univariate prediction (i.e using a currency's previous data to predict it's future movement).

The above reflects my experience and that which I have observed since the early 1990s. Many of you may have had different observations and results.

The univariate call in currencies is much stronger than the multivariate suggestion in non-currency markets.

The univariate nature of currency prediction DOES NOT HOLD for emerging market currency pairs.

Before 2005 currencies and multivariate prediction fared well, since then it has been very poor. A major change in some hidden 'Baconion Cycle' occurred then.

I am talking here about minutes to 2 days rather than anything longer.


Comments

Name

Email

Website

Speak your mind

Archives

Resources & Links

Search