Wednesday, July 20, 2011

Day-Ahead Live Cattle Futures Price Predictive Model

The evolution of markets has been particularly apparent over the past few years.  Certain markets are now evolving into their bigger/faster/stronger selves, and conventional methods of market negotiations no longer seem to work as well as they used to.

This much we know.  

Markets do not trade with the "looseness" that they used to.  This is an earmark of markets that experience increased competition.  Everyone has to fight harder for smaller margins.  If a market participant decides to throw in the towel, the bigger/faster/stronger market will quickly fill his/her void with another eager participant.  Markets miss-eth no man.

On to the point...

A market that has been rapidly evolving over the past few years has been the live cattle futures market.  What used to be a haven market for fundamental ag traders is now becoming something other than that.  Live cattle (LC) has begun a metamorphosis that has left many dumbfounded.  Like many markets, LC is being quantized/computerized.  Heavier volumes are now in the Globex contract, volatility has expanded, and LC has generally become...well...a whole new animal.

What can be done?  Can the old-school traders beat back the new bloods?  This is unlikely, as the new entrants have already permeated the market and will only leave under the guise of regulatory reforms.   The new bloods have the golden goose.  If you had the golden goose, would you give it up so easily?  I did a piece of analysis not too long ago that created a day-ahead LC price forecast based off of certain independent variables.  The model is a multiple linear regression, and the inputs were selected using a stepwise algorithm.  The data that I used to create/validate/test the data were of daily granularity, and encompassed the period 11/1985 through 04/2011.  I did not eliminate certain colinear factors, as I wanted to see how they compared in the model.  This adjustment for multicolinearity would be done via principal components analysis and subsequent dimension reduction...so get off my back.


The list of inputs includes:
1) Feeder cattle futures prices, volumes, and open interest for the prompt contract.
2) Dollar index futures prices, volumes, and open interest for the prompt contract.
3) Corn futures prices, volumes for the prompt contract.
4) Live cattle open interest (no prices, as I'm trying to predict price using other factors).


That is 9 inputs used to predict tomorrow's LC closing price.

I'm getting really tired of typing now, so I'm going to post the goods.  Point of all of this is to challenge yourself to adapt.  If you don't know where to start...I'm providing you with a gentle nudge.

Model RMSE:  6.2537

Input variablesCoefficient
Constant term20.47030067
Corn_Close0.03267086
Corn_Volume-0.00000992
LC_OI-0.00007584
D_Close-0.02621647
D_Vol-0.00047742
D_OI-0.00017357
FC_Close0.8829565
FC_VOL-0.00028675
FC_OI-0.00040499


Simply multiply all of the above beta coefficients by their respective inputs for today, add the constant term, and ponder the output (which, again, is a predictive model for tomorrow's LC price).   In the chart below, note how the differential between the model and LC actual has become smaller over the last two years.  These errors are "drifty," but look like they're tightening up (working harder for smaller opportunities...even when armed with smart bombs).

Competitive markets...gotta love 'em.

No comments:

Post a Comment