Skip to content

PODCAST – Forecasting models in rubber and the greater commodities sector

    Helixtapping the Industry
    Helixtapping the Industry
    Forecasting models in rubber and the greater commodities sector

    Arusha – In today’s world full of unprecedented setbacks, to map the true scale of opportunity, data analytics is imperative. Moreover, to gain a competitive edge in the modern world with accurate and actionable decisions every industry needs to capitalize on leveraging data, or to be more specific predictive analytics or forecasting.

    Hello and welcome to Helixtapping the industry, a series where we examine the forces driving the rubber markets today. I am Arusha Das, Head of pricing, data, and research of Helixtap Technologies.

    I am joined today by one of Helixtap’s advisors and our Lead Quantitative Strategist, Dr. Belal Baaquie.

    Arusha – Hi Dr. Baaquie.

    Prof – Hi Arusha, how are you?

    Arusha – I’m good thank you. We will be talking about predictive forecasting and modeling today with Dr. Baaquie. But to begin could you explain your transition from theoretical physics to finance and some of the commonalities you’ve observed? 

    Prof – I actually became interested in finance in 1971 when I realized that the famous Black-Scholes equation for option pricing can be recast as an equation of Quantum Mechanics. So based on this realization that one can use the mathematics of Quantum Mechanics, I have proposed mathematical models for interest rates, options, bonds, commodities and both micro and macroeconomics. I found out that all of finance and economics, especially the quantitative aspect can be actually modeled quite well using quantum mechanics – the mathematics of it. So clearly, once you make a model you really want to know is it valid for the market or it’s just a fanciful way of thinking?

    So my students and myself, we ran numerous tests over many years of my models and they turned out to be surprisingly accurate for a diverse range of instruments that I modeled and across many markets – FX, options, options on equity, options on interest rates; they are called swaptions. But I never applied them onto the market, you know as an academic I was interested in the theoretical aspect of the market and of course, its empirical testing of the models.

    The key reason why the mathematics of quantum mechanics can be used for modeling instruments in finance is because the future of these instruments is clearly uncertain and random, as is the future of economic activities and hence this uncertainty and randomness requires a kind of mathematics which is actually ideally suited if you use the mathematics of quantum mechanics for this mathematics. So the mathematics of quantum mechanics is ideally suited for explaining random phenomena and finance and economics, especially the future, falls in that domain. So that’s the reason why I persisted in modeling these phenomena with what I call quantum mathematics. 

    Arusha – That is a very interesting link. Now to talk about the proprietary predictive model – what are some of the key use cases you’ve built it for – how would you suggest market participants use this for their benefit?

    Prof – There are two regimes for forecasting commodity prices, short term – which is less than a month, and long term – a month out to a year. I mean preferably, most models in the market, they give you quarterly, which means three monthly forecast into the future, but one can actually push out to one year. Forecasting models become important for long term trends as this allows for firms to plan their cash flows and take long term positions.

    So for long term trends, it is no point that traders try to guess based on intuition and his experience of the last few weeks. It turns out in most cases that leads to wrong results. So if you really are thinking of forecasting out beyond a month, you really need some guidance from quantitative models. That’s where I find that it’s most useful. Of course there are many more other applications like if you want to hedge your position, you want to regulate your cash flow. But all of them fall into a range of instruments which are coming into play after a month or more.

    Arusha – Compared to the oil markets which have been largely financialised, where do you see rubber in this and where do you expect we will get to in the future?

    Prof – Rubber still has a long way to go to catch up with the oil markets. The value of the global oil market, which includes crude oil, biofuels, natural gas and others, in 2019 was about $1.6 trillion. There were 1.2 million futures oil contracts everyday on the Chicago Mercantile Exchange; everyday and I’m talking about 2022. Rubber production on the other hand is only about 14 million tons per year with a price of about $1,500 per ton. That gives you a total volume of rubber about $21 billion. So the oil market has a volume 50 times bigger than rubber and the oil derivatives market is very liquid and also very volatile.

    So rubber I think is in a different category of commodities than oil. The rubber market needs to develop, in my view, financial instruments such as derivatives and options for the proper hedging of open positions as well as for creating greater liquidity. My view is that derivatives, and especially options, is the direction towards which the rubber industry will develop and should develop.

    Arusha – What is something that you brought from your time as a physicist to the finance and commodities world? Is that what inspired you to create the predictive model with Helixtap?

    Prof – In physics, the future is explained in terms of correlation functions and mathematical models that explain these correlations functions. Now correlation function is something which is different from cause and effect. You know cause and effect is that when you kick a football it will fly in a certain direction in which you kicked it. In a correlation function there is no cause and effect, you have a price today, there is a likelihood that the price one week down the line will be similar or dissimilar to the price today. So looking at correlation of prices in a timeseries, so in finance the correlation functions to me are the key to any kind of predictive model. The predictive models of Helixtap we have developed are based on the idea that the price of rubber is determined by its auto-correlation between correlation with itself and the past as well as its cross-correlation with other commodities that determine the consumption of rubber. Based on this idea, we have developed a model which can predict out to even four months beyond the quarter.

    Arusha – Ok that will be interesting, you said that its not largely driven by cause and effect.   The COVID era and this year had brought about huge volatility and many black swan events. From the Russia-Ukraine crisis, China’s zero covid policy, and also most recently Pelosi’s visit to Taiwan all impacting rubber and wider commodities markets. Looking at Helixtap’s predictive models, how has the accuracy been impacted and how would users utilize the predictive tools to navigate these markets?

    Prof – Helixtap has an advantage in that its predictive models are based on seven years of proprietary historical data and continuous flow of data sets from a wide range of contributors. So this data set is the core of the advantage that Helixtap has in creating predictive models. Now as far as the black swan events go, you know the equity markets have been devastated by them, as well as the bond markets. But the commodities markets, surprisingly, have not been that adversely affected. Of course you know oil is different, it’ss political so its volatility is closely tied to the policies of big governments

    The weekly forecasts based on Helixtap’s data going through the black swan events the last few months have a relative root mean square error of less than 2%; which is quite good. The forecasting model for rubber prices has been holding through the black swan events of this year. The error for the forecasting of longer time intervals obviously gets progressively higher as you go further into the future but it is still within the bounds set by the forecasting model. The forecasting model gives you a prediction and gives you a bound – upper and lower bound – which we expect the model to work. You can’t precisely predict the price, you can predict the price plus minus some error, and that error sets the bound – the upper and lower bounds – for predictive price. So far, we have not broken out of the bounds set by the forecasting models.

    Now short term traders use their intuition and weekly and  monthly moving averages of rubber prices. For any position, be it long or short, which is more than a month, the trader needs to be guided by quantitative forecasting as I mentioned earlier. In my view, the most efficient use of forecasting is in taking appropriate futures positions and hedging these positions using options. For pricing the options, we need mathematical models.

    Arusha – Ok, that is really interesting. Could you explain your methodology a little bit to our listeners? How does this compare with more traditional econometric models?

    Prof – Alright that’s a nice question let me answer that. So as I mentioned previously, the forecasting model is based on the historical behavior of the auto-correlation of rubber prices and its cross-correlation with other carefully chosen relative commodities. Now that’s a similar approach taken by econometric models. They have an expansion of the price in terms of what they think are the primary drivers of the prices and so if you confine yourself to a fixed data set, then the econometric and correlation models yield comparable results. The reason is very simple, the reason is that the econometric model can keep on intuning more and more drivers for the prices and they can then essentially saturate the prediction using the data points we have. So it becomes comparable, I won’t say it’s the same. However, the story doesn’t end there, at that point you’re simply dealing with data, data sets and data streams, the correlation functions are computed based on the data set you have. There is nothing else coming into it.

    However, if you go further and construct a mathematical model for the evolution of commodity prices, in particular, say rubber and related commodities, then these models depend on parameters of the model which are absent in econometric models and these parameters need to be calibrated using historical data. Forecasting based on these models for which the parameters are recursively updated  – these forecasts in general, are more accurate and reliable than econometric models.

    When we do the forecasting, we obviously run the econometric model also to see how the two compare and that’s to be fair to the model. The model if it doesn’t perform the econometric model we need to know that. However, even for the case when the predictions are comparable between the mathematical model and econometric model, the mathematical model gives far more insight into the forces driving the commodity prices. Because you have a model in which you have say one or two parameters which are controlling the evolution of the commodity prices; this concept is missing in econometrics. In econometrics is all they have: you have to put down from your own imagination and your knowledge of the market, the primary drivers of the prices and if your initial guess is wrong, you’re way off and then you’ll never get the results. So in economic models, it is very important that you decide what goes into the model, but if you do mathematical modeling of the commodity prices you’ll discover they’re surprisingly almost 90% of the prices of the commodity are based on the historical price of the commodity by itself.

    The other commodities influence it approximately 10%, so 90% of the price you actually can get from the model itself; this you cannot get from the econometric approach. For the econometric approach you need to know accurately, like for example if you’re doing rubber prices and you have to link to say the production of automobiles. For example, suppose you don’t include that, then your prediction is off. You may have to include something like copper which goes into processing rubber to make tyres, for example. So if you don’t put copper into your economic model, you’re lost. So in my view, it’s not that economic models don’t work. The problem there is that you need to put in a lot of information ad-hoc, by yourself. So I prefer a mathematical model where 90% of the prediction is based on the commodity price itself. And that also makes sense because we have markets which are autonomous, you have a market for rice, you have a market for oil, you have a market for sugar. These are autonomous, other commodities of course influence them, but they influence them I would say 5% to 10% so that’s the real, in my view, difference between a mathematical model and econometrics. And I’m giving you an elaborate answer because this is often raised, I mean logically so: how are these two approaches different?

    Now to use the mathematical models, one of course needs a fairly long time series of the commodity prices for calibrating the model and calibrating the model means to determine from the market what are the parameters of the model and also as you can imagine the market has “regimes”, the market “switches”. “Switches” means the parameters undergo change also, they are not constant, so you need a fairly long time series to see how far can you go with the parameters before “regime switching” happens, and a lot of the intricacies you have to study. And so here, of course, I find that Helixtap has an advantage because we have seven years of primary data which is proprietary and not available to the others. Then you can do modeling and you can do forecasting which others can’t do and so that’s what I feel that it is the direction that Helixtap is moving towards.

    Arusha – Thank you, Dr. Baaquie. Indeed predictive analytics tools to state the probabilities of the possible outcomes in the future is the best option we have to deal with the economic volatilities. Knowing can help market participants to plan many aspects of their business.

    To find the weekly price predictions for TSR please check-out, the predictive price segment of www.data.helixtap.com.

    If you found today’s episode insightful, let us know at marketing@helixtap.com!

    For more updates on the Rubber industry, please check out www.helixtap.com, and you can also follow us on socials under the handle Helixtap.

    Thanks for tuning in to “Helixtapping the Industry”. Until next time!