Home Blogs Chemical Connections Chemical Investor Forum: Exploring regression analysis to forecast Westlake’s earnings

Chemical Investor Forum: Exploring regression analysis to forecast Westlake’s earnings

ICIS
By Daniel Fletcher-Manuel on 03-Aug-2020

The multiple linear regression analysis technique shows a 0.87 R-squared value for Westlake Chemical’s quarterly EBITDA (2012-2020) as the dependant variable against a basket of 17 ICIS chemical prices.

This suggests that in 87% of cases, Westlake’s quarterly earnings can be explained by movements in the basket of ICIS chemical prices.

Financial analysts can use this a jumping off point for two further analysis exercises. Firstly, one should explore switching out some of the lower-correlation independent variables for higher-correlation data sets to see if this increases the R-squared value. If we can achieve an R-squared value of over 0.9-0.95 then this will be highly instructive. Secondly, we must qualitatively explain why each data series in the independent variables basket (the ICIS chemical prices) is impacting the earnings. For some, it is self-evident – these prices closely align with the sales price of the chemicals in Westlake’s commercial portfolio. For others, it is less obvious. Indeed, some of the prices in the chemicals basket are for APAC – so are these prices introducing an element of ‘sentiment’ i.e. could these prices point to where Chinese industrial output rates are going in the month ahead?

Clearly, further research is needed to have total confidence in both the R-squared value and what this means for forecasting Westlake’s earnings. However, even in its current state, the model is an interesting sense check for qualitative analysis already performed ahead of Thursday’s Q2 earnings announcement.

Background

Westlake Chemical will announce its Q2 2020 earnings on 6 August. Like most industry peers, Westlake has faced pressure from the global pandemic and resulting decline in end-use demand for its chemical portfolio. However, as the below chart shows, Westlake’s valuation (blue) has been recovering through Q2, in line with physical chemical markets as shown in the ICIS Global IPEX (red).

Westlake’s business is broadly split into two – mid-stream, which manufacturers ethylene and styrene monomer, and downstream (polyethylene, vinyl and derivatives). Most of its ethylene production supplies its internal needs in its polyethylene and vinyl businesses. Certainly, polyethylene and PVC demand has been damaged by the weakened consumer markets – although, how weak those segments truly are will only be evidence one governmental stimulus packages wind down later this month. However, what is very clear is that Q2 earnings are very likely to be much worse than Q1 earnings.

Multiple Linear Regression

Formula: yt=β0+β1×1,t+β2×2,t+⋯+βkxk,t+εt,

For decades, analysts have been performing regression analysis with a view to predicting outcomes from otherwise tough to forecast data sets. Quants in the financial markets typically use regression analysis as a first port of call – however, while choosing the dependant variable is easy (valuation, EBIT, cashflow), working out what in the world of big data might be instructive (and where to even start) as an independent variable is challenging.

I did a few tests with just one independent variable but couldn’t get an R-squared of above 0.55 – and at that level you had might as well use the Dow Jones Industrial Index or the oil price. I then decided to pull together 17 ICIS chemical prices covering the products Westlake sells but also other chemicals which might best point to industrial demand on a broader level.

The output was a 0.87 R-squared value. This suggests that in 87% of cases, Westlake’s quarterly earnings can be explained by movements in the basket of ICIS chemical prices.

I would love to get your feedback on this. What should I do next? Email me at daniel.fletcher-manuel@icis.com with your suggestions or to get a copy of the model for you own use.