Buying NFT capitulation: a theorem

4 min readJul 10, 2023

Recently we have seen record-breaking NFT liquidations, with both lenders and borrowers losing money.

At some point surely the liquidations stop and price will recover to some extent. I’ve been doing some thinking — how can this be modelled, and more importantly, how can I profit off this?

I have a theorem of sorts I thought of a while ago, which may or may not already formally exist. In this article I’ll try explain it and apply it to the theory of NFT capitulation.

Let’s paint a scenario: Bored Apes are at 40Ξ, machibigbrother.eth instantly dumps 100 apes into the Blur bidding pool, and this drags the price down to 30Ξ.

Let’s assume there has been no price-changing news, and Machi is the only seller. Assume also there are 10 passionate buyers of Bored Apes. We don’t know the prices they are willing to buy at; assume they haven’t placed any bids for us to see. Assume they will all become aware of the new price at different times. A visual example is shown below.

Potential buyers at different times

By t=3, 9/10 buyers have become aware of the new price. If the price has moved up, it’s because some of these 9 buyers believe that 30Ξ is a reasonable entry point. Some might even believe 31Ξ, 32Ξ, 33Ξ, 34Ξ, are all good price points to buy. If this is the case, buying at t=0 was a smart decision since you could sell higher to these passionate buyers.

If none of the 9 buyers want to buy at 30Ξ, and you bought at 30Ξ, you can simply resell at breakeven or a small loss. We can see that Machi’s wallet has no more apes to sell. So whether or not we have buyers at 30Ξ, the overall expected value makes it a smart decision to buy Machi’s instant capitulation to 30Ξ.

Of course, this is grossly oversimplified in comparison to reality. For a larger population, and more randomness around when each buyer has noticed the new price, we could model the expected percentage of potential buyers (y-axis) that have seen the new price at time t (x-axis).

It may not necessarily be the curve of this lognormal CDF, but it illustrates the point.

At a certain time t=k, we can assume most buyers have seen the new price

At t=k we can assume price has adjusted accordingly, by either staying the same or moving up as described above. If we bought at t=0 it would make sense to sell at t=k.

We might be able to estimate the time k based on previous data, scrolling Twitter and Discord, or even just intuition.

This is still very much oversimplified, since in reality we might find new sellers at 30Ξ that were planning to hold at 40Ξ, or we might have other traders competing to buy at 30Ξ. But it is a framework that can be used, especially when it is only one wallet that is dumping the NFTs (which is often the case).

Mean-reversion is not a new phenomenon; moving averages have been used as indicators for ages. But the NFT market is still very inefficient due to a concoction of illiquidity, lack of charts, and irrational participants (this is crypto after all).

Let’s take the example of Mando (@rektmando) and OSF (@osf_rekt) dumping apes into the Blur bidding pool (February 22, 2023).

The capitulation wasn’t instant since they sold into a thick wall of Blur bids, but price spiralled shortly after they sold as bidders played hot potato to offload Apes they never wanted to buy in the first place. The graph shows a nice illustration of how it played out (pinned on the axes are the price and rough time of sale).

Another example is GCR (@GCRClassic) selling a few hundred Miladys. Again, price and rough time of sale are pinned. The bottom only a few hours after he sold was noticeable since he was selling in multiple transactions over an hour or so. Although he still had some Miladys left, the selling transactions stopped.

Anyways, food for thought, I’m sure this theorem or a variation has a name somewhere. Keep it in mind and don’t be afraid to buy ugly JPEGs to sell upon mean-reversion.