Pinned post
Brett Caughran
Verified account This account is verified. Learn more@FundamentEdge
Training Humans at Fundamental Edge
Training Agents at Fundamental Research Labs
Training Agents at Fundamental Research Labs
4.6K Following 63K Followers 4.7K Posts
Latest posts 5
Same, applied to investment research
Perplexity / Perplexity Computer isn’t my daily driver anymore, but for deep research requests like this, a well designed Skill pipeline in PC can create something really comprehensive (notably better than Claude)
Perplexity / Perplexity Computer isn’t my daily driver anymore, but for deep research requests like this, a well designed Skill pipeline in PC can create something really comprehensive (notably better than Claude)
This can be a golden age of investigative journalism. Analyzing filings, records and public data to go deep into a company or government or whatever used to be something that took a team of journalists a long time, so was relatively rare. Now LLMs can 10x investigative research.View quoted post on X
As a PM, onboarding a new, fresh junior analyst is a terrible experience. At best, it's a six-month distraction, and that junior analyst is almost always a near term cost center to the business. Workflow by workflow, you have to show that junior analyst what to do, and very closely monitor that analyst's work for errors, providing direct feedback along the way.
And that’s the easy part. The hard part is enforcing how to think. Un-training the overconfidence out of that junior analyst, and pushing back harshly when the analyst makes a claim unfounded by the primary research.
If you asked a PM what the ROI of training a junior analyst is around month four or five, without looking ahead, they would say it's terrible!! I hate this, and I never want to bring on a new junior analyst again!
Usually around month six, that analyst can then be trusted to conduct workflows on his or her own. Okay, this is a little bit better. The work product is good now. I can loosen up my oversight a bit.
And the ROI gets better.
Usually after two years, good analysts cross a threshold. And they start to become standalone insight generators, and talented stock pickers (many never make this move as some of it is innate)..That two years of feeding into that junior analyst starts to pay off, and at the end of the year, you can start to notice that analyst's impact in the P&L of the business. At this point, you're grateful for the two years you spent training that analyst, and while the junior analyst was more expendable to the business, the talented senior analyst is incredibly critical to your efforts.
I think that's where we're at now with agents.
The ROI of almost all motion on AI for investment research over the last three years has been definitively terrible for public market investors. I don't think it has really bent the alpha curve anywhere, consistently and at scale (certainly hasn't shown up in P&L at scale anywhere). And I've heard many complaints about all the hours wasted for no payoff. But also, we shouldn't expect a junior analyst to bend the alpha curve.
But a senior analyst? A well-trained, proactive, end-to-end system. That's much more interesting. That is an entity that can start to bend the alpha curve in a portfolio.
The most forward-leaning public market investors are very close to having swarms of senior analyst talent at their fingertips. My belief is that over the next 6 to 18 months, this will start to bend the alpha curve in a noticeable way.
And that’s the easy part. The hard part is enforcing how to think. Un-training the overconfidence out of that junior analyst, and pushing back harshly when the analyst makes a claim unfounded by the primary research.
If you asked a PM what the ROI of training a junior analyst is around month four or five, without looking ahead, they would say it's terrible!! I hate this, and I never want to bring on a new junior analyst again!
Usually around month six, that analyst can then be trusted to conduct workflows on his or her own. Okay, this is a little bit better. The work product is good now. I can loosen up my oversight a bit.
And the ROI gets better.
Usually after two years, good analysts cross a threshold. And they start to become standalone insight generators, and talented stock pickers (many never make this move as some of it is innate)..That two years of feeding into that junior analyst starts to pay off, and at the end of the year, you can start to notice that analyst's impact in the P&L of the business. At this point, you're grateful for the two years you spent training that analyst, and while the junior analyst was more expendable to the business, the talented senior analyst is incredibly critical to your efforts.
I think that's where we're at now with agents.
The ROI of almost all motion on AI for investment research over the last three years has been definitively terrible for public market investors. I don't think it has really bent the alpha curve anywhere, consistently and at scale (certainly hasn't shown up in P&L at scale anywhere). And I've heard many complaints about all the hours wasted for no payoff. But also, we shouldn't expect a junior analyst to bend the alpha curve.
But a senior analyst? A well-trained, proactive, end-to-end system. That's much more interesting. That is an entity that can start to bend the alpha curve in a portfolio.
The most forward-leaning public market investors are very close to having swarms of senior analyst talent at their fingertips. My belief is that over the next 6 to 18 months, this will start to bend the alpha curve in a noticeable way.
The gap between investors who’ve embraced ai (especially in research + admin) and those who haven’t is staggeringView quoted post on X
One of the best I know said 90%+ of firms work is now Ai / automated
Another sole practitioner has a whole firm run by grok bots
Era of agentic investing is here
I think it's even more fundamental than this.
For many decades in markets, long-short hedge funds could construct portfolios where their longs *on their numbers* were cheaper than their shorts *on their numbers*. And so you could buy, say, a portfolio of high-quality longs growing 15% for 15x earnings and short a group of lower-quality businesses growing 3% at 20x earnings. Investment research and market expectations were a less efficient place, and so you could truly have a differentiated view relative to both Street and market-implied.
Even better, pre-GFC when you went through challenging macro periods, those high-quality longs tended to hold up better in down markets vs the lower-quality shorts (imagine that!!!). Positioning wasn't yet deterministic. As a result, it was quite common for high-performing hedge funds to put up mid-teens long-short spreads at attractive Sharps ratios. Spread PLUS downmarket protection!!
When you construct a portfolio wedge in that manner, definitionally, duration is an advantage. Really, all you have to do is be right and wait. This is how the Tiger Cub model became ascendant.
The archetypal ideas that come to mind in this wedge box are long Google and short Sears. For so many years, you could buy Google at a low teens multiple, growing 30, and shorting Sears into the dirt. And the times when Google did pull back from 13 times to 10 times, those were phenomenal buying opportunities.
If you look at the Cao, Chen hedge fund crowded position research and the research on underperformance of crowded shorts, just owning effectively the consensus hedge fund long-short book gave you roughly eight points of spread over three decades. The top tiger funds did materially better than this.
Over the last 18 years, things have slowly changed. This started in the GFC when crowded names like Apple underperformed highly crowded, low-quality businesses with high short interest (exacerbated by poorly constructed short-selling bans). The short side in particular has become much more crowded, and the GME-AMC saga further highlighted the market microstructure challenges on the short side.
But mostly just the success of the long winners, short losers trade invited more competition. Rising competition creates more crowding. It's almost definitionally a negative to Sharpe.
So this long-running deterioration in the spread opportunity more recently has met a much more narrow leadership in public equity markets. And that leadership on the long side tends to be dominated by companies that have a very wide fan of potential intrinsic value outcomes. Either the NeoClouds or even companies like NVDA, where the current level of profitability is so far out of historical norms. These are just very clearly much harder businesses to value than Google growing 30, trading at 13.
AI in general as a theme starts to touch the terminal value perception of nearly every company in markets. If you deconstruct the net present value of nearly any stock, the majority of value today lies in the terminus or the 10 to 50+ year out collective market view of those cash flows. As an economy that's in accelerating transformation, how does the crowd know what any business looks like in thirty years? The market obviously won't. And so very clearly, the public markets will just be structurally higher vol than they ever have been before. (I think probably the consensus view on X is that AI will make markets more efficient and boring, i.e. lowering equity vol, by making investment research more accessible. I believe the exact opposite is happening and will only accelerate due to structural terminal value distrust)
What does this mean for the future of long-short equity? I don't know, but I have a couple strong opinions.
One: The “buy good companies, short weak companies”, portfolio wedge approach is probably structurally impaired. Particularly if you are trying to express it in a 250-plus GMV wrapper.
(Continued in replies)
For many decades in markets, long-short hedge funds could construct portfolios where their longs *on their numbers* were cheaper than their shorts *on their numbers*. And so you could buy, say, a portfolio of high-quality longs growing 15% for 15x earnings and short a group of lower-quality businesses growing 3% at 20x earnings. Investment research and market expectations were a less efficient place, and so you could truly have a differentiated view relative to both Street and market-implied.
Even better, pre-GFC when you went through challenging macro periods, those high-quality longs tended to hold up better in down markets vs the lower-quality shorts (imagine that!!!). Positioning wasn't yet deterministic. As a result, it was quite common for high-performing hedge funds to put up mid-teens long-short spreads at attractive Sharps ratios. Spread PLUS downmarket protection!!
When you construct a portfolio wedge in that manner, definitionally, duration is an advantage. Really, all you have to do is be right and wait. This is how the Tiger Cub model became ascendant.
The archetypal ideas that come to mind in this wedge box are long Google and short Sears. For so many years, you could buy Google at a low teens multiple, growing 30, and shorting Sears into the dirt. And the times when Google did pull back from 13 times to 10 times, those were phenomenal buying opportunities.
If you look at the Cao, Chen hedge fund crowded position research and the research on underperformance of crowded shorts, just owning effectively the consensus hedge fund long-short book gave you roughly eight points of spread over three decades. The top tiger funds did materially better than this.
Over the last 18 years, things have slowly changed. This started in the GFC when crowded names like Apple underperformed highly crowded, low-quality businesses with high short interest (exacerbated by poorly constructed short-selling bans). The short side in particular has become much more crowded, and the GME-AMC saga further highlighted the market microstructure challenges on the short side.
But mostly just the success of the long winners, short losers trade invited more competition. Rising competition creates more crowding. It's almost definitionally a negative to Sharpe.
So this long-running deterioration in the spread opportunity more recently has met a much more narrow leadership in public equity markets. And that leadership on the long side tends to be dominated by companies that have a very wide fan of potential intrinsic value outcomes. Either the NeoClouds or even companies like NVDA, where the current level of profitability is so far out of historical norms. These are just very clearly much harder businesses to value than Google growing 30, trading at 13.
AI in general as a theme starts to touch the terminal value perception of nearly every company in markets. If you deconstruct the net present value of nearly any stock, the majority of value today lies in the terminus or the 10 to 50+ year out collective market view of those cash flows. As an economy that's in accelerating transformation, how does the crowd know what any business looks like in thirty years? The market obviously won't. And so very clearly, the public markets will just be structurally higher vol than they ever have been before. (I think probably the consensus view on X is that AI will make markets more efficient and boring, i.e. lowering equity vol, by making investment research more accessible. I believe the exact opposite is happening and will only accelerate due to structural terminal value distrust)
What does this mean for the future of long-short equity? I don't know, but I have a couple strong opinions.
One: The “buy good companies, short weak companies”, portfolio wedge approach is probably structurally impaired. Particularly if you are trying to express it in a 250-plus GMV wrapper.
(Continued in replies)
Man I was really liking Instinct as a personal agent, but Claude Duet really is much better. Finally got off waiting list today. Onboarding and UX is so delightful. As I assumed, agent harness sitting on the Anthropic models is so incisive.
Now I need to head to head Grok @bot, which has been my obsession.
Which got a huge atta boy from me today for calling Apple to check on MacBook PROLED inventory (touch screen Mac book with a better camera for Zoom OMG yes 🙌🏻 )
Now I need to head to head Grok @bot, which has been my obsession.
Which got a huge atta boy from me today for calling Apple to check on MacBook PROLED inventory (touch screen Mac book with a better camera for Zoom OMG yes 🙌🏻 )
