This month’s post has three sections - AI, Investing, Miscellaneous.
Majority of content in this section is from a friend in the industry, paraphrased for clarity 
AI is supposedly slowly surpassing humans in “lower-level” cognitive tasks like speech recognition and image recognition . But whether AI can surpass humans in “higher-level” cognitive tasks like thinking of a new business idea has yet to be seen. Deep Learning might not be the solution, as it is constrained by its training dataset and can’t “transfer” easily to a separate problem domain. It can’t synthesise disparate concepts to formulate new ideas like humans do.
There is, however, exciting work on Transfer Learning, in which a Deep Learning model is trained on one task, and is then used as the basis for a second related task. Was first used in Computer Vision, and recently trialed in Natural Language Processing in 2018 with apparently amazing results. Multitask Learning, where you learn multiple tasks simultaneously, might be another approach towards a general AI, and Google’s working on this. My friend thinks it will take a combined mix of algorithms, compute power, and big data to result in step level changes in AI, not just the algorithmic improvements mentioned above .
As fascinating as successes have been, it’s also worth realising the large amount of hype and primitive methods in the industry. I like reading about the sector but find it hard to distinguish practical results vs publicity. I’m focused on this despite that because the disruption posed by AI could be incredible, if it works.
A friend forwarded me this post on investing, and it’s one of the best I’ve read. Select quotes:
DCF-investing works because most other investors also believe, or will eventually, that a DCF is a reasonable measure of fair value for a company. Investors deal in relative, not absolute truths.
I’ve read and liked Mauboussin’s writing on what a PE multiple means based on fundamental principles, and he also has an EV/EBITDA piece. However, I’m not sure how much of the buyside or sellside thinks of multiples the way Mauboussin outlines. Sure, you can take the inverse of the P/E to get earnings yield  and compare that to your return expectations, but what’s the equivalent for EV/Sales or EV/EBITDA? People put a multiple on EPS to get a price target, but how do you justify the number in a vacuum, without referring to comps? I’d argue a multiple mostly represents investor sentiment rather than fundamentals.
its not the sophistication of his process that matters, it’s the quality of his process relative to the pool of competitors – and likely his pool of competitors has changed.
Outperformance in investing comes from an edge over the market. I like the way John Huber talks about sources of edge being information, analysis, or time . Even if you have relevant information e.g. Apple iphone sales data, you can’t make excess returns if that information is public knowledge. Even having a great framework and process is irrelevant if everyone else uses the same ones at the same time. It’s only when you know ahead, analyse better, or hold longer in comparison to others that you make excess returns.
Value investors are front-running, the same way momentum investors do, the same way scalpers do in an open-outcry pit
I think most investment is front-running. Suppose you bought a $1 stock in a non-dividend paying company you think is underappreciated by the market, a “hidden gem”. If noone else ever comes to share your point of view, regardless of how good the company continues to perform, the stock won’t move and you’ll never get a return. You make money only when the new narrative of the company spreads and others come to believe in the same story you do.
most stock market strategies at their heart are parasitic exploitations of the inefficiencies of other human participants. The implication of the shifting market eco-system from human-to-machine participants means many strategies depend on implicitly assumed conditions of the habitat that may no longer be true
If your process was based on experience of 20 years in the markets, the frameworks you built in that earlier time without machines trading might no longer be helpful and could even detract from performance. Could partially explain why returns have been poorer for professional investors, though this is speculative and hard to prove.
the most successful investors are not those who succeed within a single regime, like the tribe of the value investor, but those investors who are successful across regimes
I can’t prove this but worth pondering. I also wonder, how do you train for this?
Another investing piece is this one on trying too hard:
we should be more content with probabilities and admit that we really know very little. […] expertise beyond a minimal level is of little value in forecasting change […] No matter how much evidence that seers do not exist, suckers will pay for the existence of seers.
Although I agree that we know little and forecasting is nearly impossible, I think the reality is that people pay you to act like you know and to forecast. Unfortunately I think the upside of having a strong opinion, regardless of fundamental soundness, overrides the advice of admitting we know less than we think. Let me know if you disagree, but there seems to be an incentive to err on the side of provocation.
Lack of clarity is especially helpful when content is poor. Because there’s maybe two billion dollars in fees at stake in our business, we can be tempted to avoid simple investment methods, which can be simply explained to our clients.
I don’t know if true but sounds reasonable. People seem to think the more complicated the better, be it investment strategies or email newsletters littered with non-hyperlinking footnotes and a miscellaneous section that get longer each month.
Speaking of being persuasive, here are some tips on being more impactful even if your job is mostly analytical work.
Because it’s rare, the ability to excel at both logical reasoning and relationship building is especially valuable. One trick is to get to know people before you need them.
Related to what I’ve written about before regarding become the top percentile in two or more areas rather than just one alone.
We might like to think a model, an idea, or a data set should stand on its own, but that is simply not how people are persuaded. They care about the persuader. They need to believe in the person selling the idea. […] You need evidence of personal familiarity with the front lines — stories, contacts, concrete examples. Find ways to accumulate that evidence.
Facts alone don’t win arguments, but you need to sell the story to your audience. Selling is such an important skill that is also underappreciated by many. I took a sales class in college before, and still wish I had learnt and practised more . For many careers, the higher up you go the more sales oriented your role becomes e.g. banking, consulting, software.
many staffers were understandably reluctant to adopt the model. To ease their concerns, we gave them veto power over every decision, and we left 10% of the class to be filled outside the model, using whatever method they desired
Interesting solution of how to get people on board with new processes by giving them some choice in the matter. Sort of like how making organ donation opt out rather than opt in improves overall outcomes for the system while still giving people freedom .
I did not finish reading this, but here’s a history of investment banking for anyone interested. 
“females on average experienced steeper increases in depressive symptoms than males over their teenage and adolescent years until around the age of 20” and how peak depressive symptoms in girls occur earlier than in boys
“I didn’t have to stop and immediately follow Isaiah’s injunction to Hezekiah - set thine house in order for thou shalt die, and not live. I would have time to think, to plan, and to fight.” Stephen Jay Gould’s experience on looking up stats for his cancer diagnosis.
Alternative take on tech antitrust sent by a friend. I haven’t consolidated my thoughts on the increased desire for tech regulation yet, but generally believe we shouldn’t regulate for the sake of regulating, and regulation has high chance of helping incumbents e.g. GDPR.
Yes, people actually do email me to discuss things, and there is more than just you reading this newsletter.
How hard can it be to differentiate a puppy vs a bagel? Rather hard, as it turns out
Algorithms referring to deep learning etc, compute power such as quantum computing, and big data such as imagenet. Each area has its own advances and challenges at the moment.
For those less familiar, e.g. a stock price of $20 on estimated 2020E EPS of $1 implies a 20x PE multiple. The inverse, 1/20, implies a 5% yield on the capital you’ve invested, roughly speaking. There’s an inconsistency in WASO vs DSO but we’ll ignore that for now. I think use of multiples is generally sloppy but have no idea how to resolve it. There’s too much inconsistency as to whether someone is using GAAP vs non-GAAP multiples, NTM vs 1 year forward, what street consensus is, that comparisons between companies are highly imprecise.
Some would add behavioural to this list as well.
I am so horrible at negotiating I leave the negotiation making both parties feel they’ve gotten the worse end of the deal…
Another study disputes whether this actually results in more donations in reality after accounting for family intervention post the deaths.
I’m still trying to find out when the term ‘merchant bank’ switched to ‘investment bank’; if anyone knows please reach out