-- John Kemp is a Reuters market analyst. The views expressed are his own --
By John Kemp
LONDON, June 28 (Reuters) - There is no evidence anyone can successfully predict commodity prices. Most forecasts seem to be adaptive -- reacting to past price changes -- rather than forward-looking, and therefore miss turning points.
There is no evidence any forecasters consistently get it more right than wrong. Forecasts can never be more than a baseline for planning and investing surrounded by significant uncertainty.
Nonethless, preparing forecasts remains a useful exercise because it forces analysts to identify factors driving prices and how the balance might change in future. It is the process rather than the outcome that is valuable.
While equity investors have embraced this approach -- valuing research for the supporting information and analysis of sensitivities as much as the price call -- too much research in commodities still emphasises point forecasts . There is little consideration of the distribution of risks or of factors that might drive alternative outcomes.
Enormous resources are now being poured into commodity price forecasting but a thoughtful paper by the New York Fed casts doubt on whether it is likely to be successful.
The challenge is to transform the purpose and presentation of research. Rather than trying to help investors and hedgers predict the future (which is doomed to failure) research should help them accept and manage uncertainty in the most appropriate way .
In a 2010 working paper on "Commodity prices, commodity currencies, and global economic developments" and now featured on the New York Fed's website, Jan Groen and Paolo Pesenti ask "how easy it to forecast commodity prices?" (www.nber.org/papers/w15743).
The authors conclude “across our range of commodity price indices we are not able to generate out-of-sample forecasts that, on average, are systematically more accurate than predictions based on a random walk or autoregressive specifications”. In other words, forecasts could not beat a prediction based on the current price or some weighted average of past prices.
Groen and Pesenti considered four families of indices (Reuters/Jefferies CRB index , Standard and Poor’s GSCI <.SPGSCITR > , Dow Jones-UBS , and IMF Non-Fuel Commodity Prices Index) and three ways of forecasting them.
The first approach focused on exchange rates from a small group of commodity exporting countries which derive a substantial portion of output and export earnings from primary products but are too small to exert monopoly power on prices and have free-floating exchange rates (Canada, Australia, New Zealand, Chile and South Africa).
The second and third added a wide range of other theoretically important factors -- including macroeconomic variables (industrial production, business and consumer confidence, retail sales, unemployment, core inflation, interest rates), commodity fundamentals (inventories, production), and the Baltic Dry Index .
Groen and Pesenti examined the out-of-sample forecasting power of these approaches over five time horizons ranging from 1 month and 3 months to 2 years and compared them with benchmark predictions based on a random walk or an autoregressive process.
For some approaches over some time periods the forecasts performed a little better than predictions based on the benchmarks, but the outperformance was fairly random. Some approaches worked for some indices at some horizons, but not for other indices at other time frames.
“Neither the exchange rate approach nor a broader approach that uses information from larger data sets including both exchange rates and other macrovariables are overwhelmingly successful in predicting commodity price dynamics,” observe the authors.
Perhaps poor performance of exchange-rate and macroeconomic based forecasts was due to the dominance of commodity specific factors in determining returns? The authors therefore consider whether including futures and forward prices for commodities, which encapsulate expectations about specific factors, could improve the predictions.
They warn “futures prices provide, at best, highly noisy signals about future price spot prices” and “ it is unclear whether prices in relatively illiquid segments of the futures market such as longer-dated contracts can be considered unbiased and effective aggregators of information”. In any event, adding futures and forward prices into the models did not improve their performance substantially.
We are left with the humbling conclusion that our ability to forecast commodity prices is no greater than our ability to forecast other aspects of the future. It will not surprise long-standing professionals in commodity markets; the authors have provided statistical proof of what most market participants and observers have long known.
For intermediaries such as dealers and even hedge funds the most successful and stable long-term strategy has been to delink profitability from price forecasts and directional trades and find other sources of added-value by providing fee-earning services such as market-making, brokerage, warehousing, marketing and logistics.
For example, Goldman Sachs has moved into metals warehousing with the acquisition of Metro, while Glencore has been keen to stress its revenues are not based on directional speculation but exploitation of arbitrages and the informational advantages and optionality that comes with trading around a huge network of physical assets.
If the market’s biggest and most successful participants cannot guarantee to make money based on (directional) price forecasts, there is no more hope for smaller investors and producers and consumers engaged in hedging, even if they are armed with the best forecasts.
For commodity producers and consumers, the challenge is instead to understand the whole range of risks they face and the most effective way of managing them, recognising that some strategies merely transform one risk into another, and in some instances it may be more cost-effective to accept the risks rather than try to offload them.
For investors, the challenge is to recognise the limit of forecasting but extract an appropriate fee for assuming risks other market participants do not want to shoulder, recognising that in most cases high returns are obtaining by accepting the possibility of large losses.
In this context, research is most effective when it helps investors and hedgers explore the full range of factors influencing prices and the range of possible outcomes, not just the central tendency but the whole distribution.
Given the proved limits of price prediction, research may be most useful when it examines in depth the full range of upside and downside risks and expresses outcomes as probability distributions. (Editing by Anthony Barker)