Market review

 

There are two very notable elements to how the market performed this quarter. The first is its behavior in the aftermath of the Iran war and closing of the straights of Hormuz, which did not have the impact one might have expected (either on the internal market dynamics or the real economy). For example, Europe Aerospace and Defense stocks underperformed by 3% over the two month period straddling the declaration of war. Contrast that to the start of the Ukraine conflict in 2022, where defense stocks were up over 50% relative to the market on the same basis.  There are some obvious, valuation-based reasons for this, but still.

The second is the way in which the dominant theme of the last 3 years – AI – reasserted itself after a fairly severe relative wobble in the first couple of months. Both these elements feel surprising, because the imbalances in the market appear all too obvious to us, and because as we set out in our last note, the returns companies can expect from the planned outsized AI investments to be paltry, which means those investments themselves are unlikely to come to pass as planned over the coming decade.

I think the two are linked. So dominant has the AI / tech investment theme become that no other parts of the market remain unscathed. By this I mean that investment in the thematic is actively sucking capital away from other areas, rather than those non-AI stocks just not joining the party. Reflexivity is real of course, so corporates respond by talking up their (often genuine but minor) exposure, meaning less capital left for the rest of the market where companies can’t or don’t make those AI related exposure claims.

In prior episodes the heterogeneity of market participants would have acted as a dampening effect on this, bringing some sense to proceedings. However, these days so much capital is allocated on a non-discretionary and or non-valuation based approach – passive, momentum, corporate buy-back, retail investors, asset-liability matching, performance chasing from allocators, narrative or meme stocks, crypto – that this isn’t occurring. Who knows when it will end, but end it will, and the odds of it producing a disruptive and disorderly event are increasing the longer the bubble continues to inflate.

We shall keep looking for out-of-favour and undervalued compartments of the market where positive change is occurring, with the objective to compound a good absolute return over the medium term. At the moment that means more than ever, eschewing the most highly valued and (currently best performing) compartments of the market.

 

Portfolio Review

 

In our launch note we discussed how using an absolute return mindset – given the likely medium-term challenge to both valuations and earnings growth for the broad market – would be a necessary approach to delivering good medium term investor outcomes.

This assumes, of course, that markets may not advance as they did in previous years. That view underpins many of Goodhart’s approaches. If a particular style were expected to consistently lead a rising market—as was broadly the case from 2010 to 2022—the logical response would be to align with that factor. This is not what we believe, and not how we build our portfolios.

In this context then we have to accept that in periods where the market is indeed led by a particular style that we believe will not offer a positive return over the medium term, we are likely to underperform.

This was the case in the second quarter for the reasons described in the opening section, although our absolute return was flat. As we have discussed previously and with increasing conviction, we don’t think in aggregate the leadership style (AI stocks and the associated, increasingly broad, supply chain) can possibly offer long-term good returns. These stocks are perhaps 40% of the market. As a summary, the reasons I think this are as follows:

1. There is no plausible way for the huge planned aggregate capex to generate enough profit to make a good investment return, except under the most extreme assumptions, such as assuming it takes over a significant fraction of the entire global corporate staffing cost base. This means that it won’t ultimately take place.

   
   

2. The low hanging fruit of AI improvement has been done. Leading edge models are trained on a significant proportion of the available data in the world and while still improving, the rate of improvement, appears to be slowing.

   
   

3. It is becoming increasingly apparent to me that there are many areas in which AI isn’t such a big improvement on human + computer (and of course there are many that are).

   
   

   I am not an AI mathematician but this isn’t a surprise as AI is, at it’s core, a mathematical function trying to find the minimum of a many dimensional surface (via gradient decent) on a vast scale. It also needs to ‘kick’ or ‘jump’ occasionally to try and ensure the minimum it is in is nearer the global minimum than the local minimum it is in.

   
   

   Intuitively then, the more a problem can be defined as a (many-dimensional) surface, the better AI would be at this (i.e. at predicting the next token within that category of problem). Chess, legal work, coding all seem to fall into this category. Recognising characters, simple drawing, solving mathematical problems, search also seem to fit reasonably well.

   
   

   But the areas where this isn’t the case such as story writing, financial modelling (in my experience), making creative leaps, novel predictions based on first principles, AI seems to struggle to match human creativity (or human + computer work).

   
   

   It also – as would be expected from a non-deterministic system – isn’t sufficiently and consistently correct, and can’t easily self-determine when it has made a mistake. So, where a lot of 100% accurate assumptions are needed for a robust determination, AI might struggle.

4. The demand shock has led to super-normal peak profits in many aspects of the supply chain. In classic capital cycle manner, there is now significant investment  into these areas. Memory demand is a key example here. Historical net margin c.5%. current net income margin: 40%.

   
   

5. Valuations

   
   

6. Ability of alternative technological approaches to make the same determinations vs the compute heavy current approach. This would render much of the expected infrastructure & capacity obsolete. For example, Small Language Models, brain-based system such as Liquid Neural Networks or

   agent swarms.  This is a railways vs canals analogue.

 

We know that many investors are consciously or unconsciously capitulating into this compartment of the market feeding, its current virtuous circle. We certainly will not be doing this. We currently have no ‘first round’ AI stocks. Hopefully we have very little ‘second round’ exposure – very underweight financials for example. We think deliberately avoiding this area and delivering the return commensurate with a portfolio of generally good businesses on 10.5x 2028 PE, for double digit earnings growth, that are positively inflecting, will provide better outcomes than jumping on the bandwagon.

It is fair to also observe that we are still a little disappointed with our returns in the second quarter. This is nothing to do with the fact that we underperformed an AI-led market, but because several of our larger positions didn’t do as well as hoped. I would highlight Cicor, Bakkafrost and Royal Unibrew.

Cicor I think is being very misunderstood by the market. They are consolidating the fragmented printed circuit board market across Europe and have about 25% of their topline exposed to defense which gives them a visible mid-to-high single digit growth over the medium term. After a large acquisition of a break-even company last year, unsurprisingly, their margin isn’t making aggregate progress this year. Couple with some understandable supply chain disruptions around memory and the Iran war, and the company has had a tricky start. The issues should dissipate over the next 24 months, leaving the shares looking far too cheap for the growth.

Bakkafrost is a long-held position in many of our portfolios, with recovery in Scotland being the main thesis. This is tracking to plan, but both the feed business and salmon price have been affected over the last 12 months by the recovery of the Norwegian salmon industry (Bakkafrost is mainly Faroese). This one-off increase in supply is coming to an end and the typical demand growth for salmon consumption of 7% should start to be reflected in prices forthwith. El Nino could quite possibly cripple both feed and salmon production in the southern hemisphere which would see an outsized improvement for Bakkafrost’s P&L from here.

Unibrew is probably the most fundamentally justified underperformance. Having distributed Pepsi for 50 years in their home markets, and done a good job, they have very surprisingly (to me!) lost the contract. This is around a 15% hit to profits in 2029 against prior expectations with growth recommending thereafter. It is worth holding the position after a 25% fall in the share price, because over the medium term they can replace these volumes with own brand production, but that will take time.

Elsewhere we have added a position in Gaztransport et Technigaz (16x FY1 PE), who have a global monopoly on LNG membranes on the basis demand is likely to remain robust, and the new management can develop a sizable service and aftermarket business from its current low level. We bought Rheinmetall after a halving of the share price culminating in a 20% fall in the loss of a frigate contract. It is now on 18x 2027 earnings. Expectations are for a 200% increase of revenue between this year and 2030. Even if that is 100% too optimistic (and remember these are long term, visible contracts), it will still be cheap.

Other new positions are Wolters Kluwer, Bonesupport, Raysearch and Scout24 which are not really AI losers in our view but have been hit on that basis during the SaaS apocalypse.