|
Post by Byron on Jul 23, 2012 17:21:35 GMT
Will, I have some questions for you. I have recently entered into the foray of commodities markets this summer. Although I have taken a basic economics course and loved it, I am an engineering student without significant background in the subject. I am currently trying to create several models to track and predict price changes in certain commodities. I am working mostly with metals and some energy commodities.
My questions for you are, where do I get started? Do you have any suggested reading? Should I study up on econometrics or is there a better way? I would appreciate any comments in regards to these questions.
|
|
|
Post by willsmith on Jul 24, 2012 15:10:47 GMT
Firstly the "certain commodities". They all work differently but broad groups can be identified.
Predicting price changes - obviously if everybody could do that we'd all be rich, and by the efficient market theory, tomorrow's predicted price change would have just happened today.
There are two types of pricing models - structural models where you identify loads of possible explanatory factors, and price, and run regressions over several (many?) years looking for which explanatory factors seem to influence price. Essentially, a glorified regression.
Then there are the reduced form models which mainly use randomness to propose many different possible future prices. Using something like monte carlo you can take the average to get your central prediction, and examine the variance in your predictions to determine how accurate you think it is.
Metals are strongly influenced by inventory as well as supply and demand. The people who know detailed future supply and demand sell their reports for big $ : World Metal Statistics and CRU.
Energy : it's easier to get public data. The BP statistical review of world energy has history (but no projections), the many reports coming out of the US EIA make predictions. For example, check out "STEO" (short term energy outlook).
Once you think you have predicted price changes, if you want to know how accurately you've done that, and whether your strategy would have made you money, there's (unfortunately) a lot more to research : data snooping, backtesting strategies, roll yield (if you are in futures), different measures of measuring success of a strategy : sharpe ratio, maximum drawdown, ...
Personally I recommend my PhD supervisor's book (Helyette Geman, Commodity and Commodity Derivatives 2005) as a good background on commodities and pricing models. If you fancy a light read, try the various books of Jim Rogers. For more core stuff, there are many books on asset pricing but they usually focus on equitites and bonds.
Personally I think a basic knowledge of econometrics is incredibly useful, but people sometimes take it too far rather than use common sense. For example, there are 100's of papers published trying to establish, using "Granger Causality" and "Vector Autorregressions" whether GDP growth causes electricity consumption growth, or electricity consumption growth 'causes' GDP growth. Kind of chicken and egg in my opinion. They also sometimes forget that it's much quicker to measure changes in electricity consumption (known instantly) than GDP (takes a few months).
|
|
|
Post by arabesca on Nov 23, 2012 22:03:05 GMT
Dear Will, how would you recommend to approach fair value modelling of agricultural prices? Reduced form or general equillibrium?
Surprisingly, i havnt found much material on this. I was trying to model corN prices as a function of quarterly US stocks, taking data from 1980. the relationship is not very significant and the explanatory power is low is is because US is not any more a dominant market maker or my model is wrong?
would appreciate your feedback
|
|
|
Post by edouard on Nov 23, 2012 22:28:23 GMT
Hello.
In practice, some market research teams in big trading houses (you see, the ones based in Geneva ...) develop structural models to forecast prices of agricultural commodities.
In practice, ag market analyst, better use stock/use ratios than stock absolute values to 'feel' whether ag commodities are over/undervalued.
Did you read the paper Soybean Inventory and Forward Curve Dynamics (Hélyette Geman and Vu-Nhat Nguyen)? It gives hints on how to use quarterly data of the USDA, and briefly discusses major recent shifts in the international trade of soybean. Maybe, that can help for your task at hand.
Anyway, which other variables did you test in your model?
Best, Édouard
|
|
|
Post by arabesca on Nov 24, 2012 7:49:23 GMT
I am indeed in Geneva . I will read this paper, thanks. I usee stocks to use, crude oil, dxy. Surprisingly stocks to use of soybeans were not significant. Are u involved in similar research?
|
|
|
Post by edouard on Nov 24, 2012 18:05:20 GMT
I sent you out a PM.
So, the models you research on are multi-regression models?
Personally, I am not satisfied with such models (and I sometimes fell lonely with my opinion).
First, one is soon confronted with multicollinearity problems with such models. Take the independant variables you mentionned here: isn't there some correlation between, USD, USD index and oil prices?.
Second, building a price forecasting model using daily, weekly or even monthly prices for on part, and quarterly data on the other hand is a bit dubious (ok. there exist statistical techniques that do the trick). The price modeller takes the quarterly data of the USDA for reason of avaibilities. It is true, USDA's estimates are essential benchmarks for market analysts, traders and other market participants. But, because of the so called 'weather-market', export bans or other major events between 2 USDA's reports, the expectations of market participants on the final figures might well have changed, and the current prices being actually reflecting those modified expectations. (I am stopping here, for the sake of clarity, but, one can add the fact that there are other major benchmarks, say private analysts such as Informa, Allendale ...). Finally, the crucial point is that quarterly (data by their very nature) forget the price dynamics.
Best, Édouard
|
|
|
Post by aqoxekahiwu on Jun 9, 2019 6:17:32 GMT
|
|
|
Post by aqoxekahiwu on Jun 9, 2019 6:18:41 GMT
|
|
|
Post by iokuxel on Jun 9, 2019 8:07:53 GMT
|
|
|
Post by iokuxel on Jun 9, 2019 8:08:47 GMT
|
|