Streetwise Professor

Today’s Bloomberg Odd Lots has an interesting conversation with Don Wilson, founder of DRW whom Odd Lots calls the “King of Chicago Trading.”

Overall the ~40 minute conversation was very interesting. One of the most interesting parts related to the possibility of GPU futures trading. Wilson thinks it could become the world’s biggest market, and is putting his money where his mouth is. This includes the creation of GPU price indices.

I’m pretty skeptical on the prospects for this market. I note at the outset that it is very difficult to predict which futures contracts will succeed, and which will not. Most don’t.

In a nutshell, the necessary conditions for a successful futures contract are pretty well established. The underlying market must be large. The price of the underlying must be volatile. The underlying can be heterogeneous, but not too heterogeneous, otherwise basis risk is too big with any single contract, and no single contract can generate enough volume to be viable.

GPU would certainly tick the market size box. Perhaps the volatility box. The heterogeneity box–not so sure, although trading a contract that prices based on a service flow (e.g., flops) rather than chips themselves would probably tick that box as well.

Although the necessary conditions for a successful contract are pretty well established, sufficient conditions are not. One can launch a futures contract on something volatile with a large underlying and price volatility and watch it flop. That happens quite frequently.

What’s missing from most analyses, in my view, are more qualitative things. The most notable of these is market structure. In particular long marketing chains, with a lot of intermediation. Also, relatively low concentration on either the producer or the consumer side.

In commodities, for example, trading firms generate a lot of the hedging activity. They intermediate the flow of the physical, and typically their exposures turn over relatively rapidly, meaning that they are putting on and taking off positions frequently. This generates trading volume and liquidity.

The fragmentation of the value chain and fundamental volatility militate against long term contracts for most commodities. Moreover, the lack of concentration mitigates market power which is often associated with price rigidity (as Dennis Carlton showed many moons ago).

My understanding of the GPU market is that the marketing chains are short, not intermediated, and highly concentrated. For example, there are a small number of very large cloud storage providers. AI is likely to be highly concentrated on the buy side. The GPU production side is obviously very concentrated, being dominated by NVIDIA.

Moreover, transactions cost considerations are likely to lead to a good deal of vertical integration. AI firms are going to buy their own chips. Vertical integration also militates against the development of a liquid futures market.

Given the concentration on both sides of the market, I anticipate that to the extent buyers and sellers want to achieve some price stability, it will be via bilateral contracts between the producers and the big users.

Joe Weisenthal asked the question that came to my mind–what about DRAM futures, which were highly touted at the height of the dotcom boom, and crashed along with NASDAQ in 2000?

I have some personal history here. One of the (many) nascent DRAM futures markets (Buckaroo.com, if you are wondering) flew me out to Mountain View to give my opinion on the prospects for the market. Ironically, I was there the day NASDAQ crashed. (Hmmm, I also had some connection with the ’87 Crash–maybe it’s me!)

My assessment was negative. Not long after, the late CEO of Micron, Steve Appleton had me come out to Boise to give my view. Buckaroo had been importuning him, and when they told him they had talked to me, he invited me to tell him what I had told them. After hearing my take, he told Buckaroo “hard pass.”

My prediction, vindicated by experience, was based on market structure considerations. My skepticism about GPU futures is similarly grounded.

For his part, Wilson dismissed DRAM as a precedent. In his view, the fact that DRAM prices trend down strongly meant that there was no underlying volatility to hedge or trade around.

I don’t find that compelling. Yes, everybody knew that prices for a particular chip type tended to trend down (due mainly to learning-by-doing). But as long as there is variation around that trend, there is a price risk that sellers and/or buyers might want to hedge.

DRAM might not be the only cautionary tale. I’m so old I remember Enron touting bandwidth as the future world’s biggest commodity. Yeah. No.

Another issue is information flow. A high rate of information flow causes high rates of price changes which drive both hedging and speculation. Weather derivatives tick a lot of the necessary condition boxes, but information flow is very lumpy. Consider a weather derivative on July 2026 cooling degree days in Houston. From now until, oh late-June–maybe–there will be relatively little information that flows. So there’s nothing really to trade on until then. Meaning that volumes–and liquidity–will be moribund.

Which is why exchange-traded weather derivatives have never done any real volume.

My intuition is that the information flow about future DRAM value will be pretty modest.

So maybe GPUs will be unlike DRAM and bandwidth and weather, maybe it won’t. I will watch with interest.

Another interesting part of the Odd Lots conversation related to the CME’s move to the cloud. Wilson made the great point that the way the cloud works is not compatible with housing the matching engine there. Especially in a high speed world, it is nigh on impossible to implement time priority on the cloud.

Wilson, Alloway, and Weisenthal also discussed tokenization. Here I agree that this could become a thing, and a big one. It may finally be the moment for DLT/blockchain technology to have a big impact.

It was widely hyped for years as transformative, for everything–including financial markets and commodity markets. I had a 2022 JACF paper expressing skepticism about the potential of DLT/blockchain. I still am skeptical about a lot of the proposed uses (e.g., tokenizing shares of fine art or real estate, or in physical commodity trading). However, tokenizing standard securities, derivatives, or collateral offers some advantages. Like I wrote in the 2022 piece, the interesting thing to watch will be whether the blockchains will be permissioned or public.

Some tokenized ETFs and MFAs trade on permissioned chains, but some are on Ether. In 2022 my intuition was that most trading applications would be via permissioned blockchains, but now I’m not so sure. I’ll revisit that issue as time allows me to get more up to speed on what is going on.