How to use AI to find trade setups before the rest of the market

AI has helped me with one part of trading I used to do manually: figuring out why something moved.

How does this contribute to returns, you may ask?

If you can explain the driver, you can look for the same trade setup somewhere else.

And AI can help you explain the "why" faster vs the rest of the market...

So this week I will show you a real case study on how I used AI for a trade I put on the books that's returned me ~35% so far...

The Background Context

As many of you know, I mostly trade crypto and crypto equities.

That is where I made most of my money doing prop trading and the core strategy for my fund I previously ran.

One token that caught my eye, ZCash, has done a 27x in the last 12 months and has been pumping aggressively last few weeks...

The natural question I asked was what was driving this outperformance...and if I can understand the drivers I can scout for other trade setups.

My leading hypothesis was that there was a new ETF conversion that was on the docket that contributed to flows.

The thesis was:

• ETFs expand access to ZCash pool of buyers to traditional brokerage accounts
• ETF structure actually removes direct spot supply from market vs trust or other more opaque structures
• Tight control over supply can kick off price flywheel garnering more attention

So using Claude Code, I am able to test this thesis to see what the timeline was, whether the flows thesis holds up, and understand if there are other contributing factors to this pump.

AND....if my hypothesis is true I can scrape for other tokens with similar ETF approval conversions that are launching soon.

The Claude Code Prompt

The important part was the research logic.

I did not ask Claude Code to prove the ETF thesis. I asked it to try to break it.

The prompt logic was:

• Rebuild the timeline from SEC filings and daily prices.
• Separate the ZEC rally from the trust-discount trade.
• Track the coins: what was already in the trust, what arrived in kind, and which flows required market buying.
• Compare ZEC with Bitcoin, Ethereum, and other coins.
• Treat the ETF as a supply sink only when new demand forces someone to source coins from the market.
• Search for other reasons that may have contributed to this price action beyond ETF flows
• Apply the same test to pending ETFs: path to launch, issuer reach, likely new buyers, available supply, and the clearest reason the trade could fail.

One prompting tip: I'm a huge proponent of framing your Claude / Codex prompts before you run them. Try writing out your steel man before writing your prompt. A steel man should contain the framework similar to how I structured it above.

The Research Findings

Full output: Link​

The AI research was able to produce (1) a backtest if this thesis was true (2) time series of events (3) other opportunities with similar setups

Let's dive into the findings!

The series of events

To explain price action, having an annotated chart is a tip I find very helpful:

Here's what happened:

• Grayscale (the ETF issuer) announced ETF trading was expected to begin around August 25.
• Registration was effective on August 24.
• ZEC rose about 46% in the four days before trading began.
• It fell about 7% on launch day, then rose another 91% through September 17.

The timing fits an anticipation trade.

Granted, the research pointed out that other coins (incl. privacy narrative coins) rose during this window too but ZCash had more ammo for an institutional bid.

How much supply did the ETF absorb?

As the research showed, the ETF wrapper for ZCash changed who could buy it.

The ETF made ZEC exposure easier to buy through brokerage and adviser channels. That gave the Zcash narrative a much larger surface area.

So how much did the ETF absorb?

Lucky for us, Claude can scrape raw primary docs and answer this for us:

On September 8, Grayscale contributed worth close to $100 million in ZCash. And once live, the ETF attracted more than $70 million of additional inflows.

This is where the ETF gave the rally more fuel.

In crypto, the spot market is usually thinner than it looks. Which means a "small" supply sink of ~9 figures can really pump price especially as shorts get squeezed on perpetual futures.

Net creations pull coins into fund custody. More buyers are then competing for a smaller pool of available spot on exchanges / on chain.

That combination helps explain the move: anticipation before launch, a wider buyer base after launch, and more ZEC sitting inside the fund taking away available supply from exchanges.

The privacy-coin rally added fuel, but the ETF was one of the engines confirming the "why" behind the pump.

Turning the Zcash “why” into the next trade

The great thing about AI is that once we break down why a trade worked, we can reuse that why as a filter.

Zcash gave me two setups to search for:

• A new wrapper that opens the token to more investors and starts absorbing spot.
• A trust conversion where anticipation closes the discount before new ETF flows arrive.

I pointed that filter at the pending ETF docket to scrape for trade setups that were structurally similar to the ZCash pump...

NEAR: the new-demand setup

​Bitwise and Grayscale both have NEAR filings in late stages of filing pricesses.

That gives NEAR two established issuers capable of putting the token in front of U.S. brokerage and adviser capital.

A new NEAR ETF starts by building its asset base. Net creations add NEAR to custody. That is the part of the Zcash trade I want to replicate: a new wrapper, a larger buyer base, and fund holdings that keep growing.

On top of that, AI suggested NEAR had similar privacy narrative that contributed to ZCash's price action.

This was a trade I put on before the recent parabolic move:

If you want to read the full research work, download it here.

It has the complete ETF screen, the filing work, and the other token dossiers.

And of course you can use the same operating principles to point this AI research strategy towards other investment styles.

Happy hunting!!

- Dave

Personal

I'm speaking in Singapore again!

I'll be hosting an AI in Corporate Earnings panel .... and showing backtests on what is signal vs noise on corporate AI initiatives.

Are a CEO's AI initiatives real? Or just lip service? How can we tell?

If you're in SG October 23 it would be great to meet some of you there in person.

Thank you to the CFA // Institute of Valuers and Appraisers // International Valuation Standards Council for hosting this one.