It's interesting to crawl through some of the entries from leaderboard people. It seems like most of their bets are neutral or slightly random, but then they'll be propped up with a lucky call on GME from January or some other single-stock pick.
For example, this poster apparently just spammed GME posts during the run-up and got lucky, but if you followed their other calls (REAL and AI) you'd be deep in the red. Their GME spam puts them in the top 3 medium traders with a 236% 1-month return, but everything else they've posted has performed terribly: https://duedilly.io/trader/r.Ottikarottiii/
It would almost be more helpful if there was an option to subtract out the well-known meme stocks like GME. Some of these leaderboard entries have not so great results for their own calls, but they happened to post about GME at just the right time to dominate their average.
Thanks! I know exactly what you mean. I've experimented with using medians instead of means for aggregating performance as well to lessen the impact of that sort of thing.
Currently, it's pretty difficult to crack the top without something like that. At the same time, I think (as another commenter pointed out) the distribution of returns for the way these traders pick stocks is pretty skewed/lottery-like. One hesitation I have about the median is that although it's more robust to outliers its less reflective of the true performance achieved.
Maybe putting some stocks in a "meme tier" and then excluding those might capture what your describing, thanks for the feedback!
Seconding the parent commenter’s request! No idea what’s easy or hard for you to implement in terms of UI or functionality, but it would be great to be able to exclude either the top N most-mentioned stocks (via a slider, with text listing them) or specific stocks by entering their symbols.
Regardless of whether you end up doing this, great work and thank you!
I don't know who you include in "most people", but from what I understand of the professionals, that's not at all how making money works. If what you describe happens, it means the portfolio risk dial is set way too high.
The mistaken idea behind that is that people think risk and reward covaries. It does for single wagers, but when assessing profitability of investments, we must look at long term growth. When we do that, there's a risk level that maximises growth.
Lower risk than that gives you slower growth but with less variability. Higher risk is just stupid because it gives you slower growth and increases your risk of ruin (by getting variability over levels supported by your capital.)
----
People who actually make money on this are good at inventing their own insurance in ways that lowers their risk significantly. They make very specific, relatively independent bets that aren't just proxies for "exposure to the market".
The Reddit people whose posts are being used here, but I can't be sure it's just a guess. I'm not talking about fund managers that follow strict rebalancing and risk guidelines. The creator of this webapp acknowledges this and says that it's very hard to be at the top of the rankings without this happening.
You're talking about the efficient frontier from Modern Portfolio Theory, no?
I'm referring to the analogous concept from E log X optimisation. (The "Kelly criterion".)
The efficient frontier based on Markowitz' mean--variance optimisation does not help you find a risk level that maximises growth, because it looks at one investment in isolation, not sequential reinvestment over a long period.
The analogous concept I'm referring to is the set of portfolios that are linear combinations of risk-free assets and the E log X optimal portfolio. These are all Kelly optimal under more and more restrictive constraints on allowed variation, but come with slower growth as a trade-off.
To reiterate, the E log X "efficient frontier" talks about how fast your portfolio will grow over time as you reinvest your gains. The Markowitz efficient frontier tells you something about the statistical nature of a single investment opportunit.
Sure, it does. But it does so in the same sense that MPT requires known mean and variance. In either case, we go with estimations.
In many practical cases, the joint distribution of outcomes is easier to estimate safely than the variance. The joint distribution of outcomes can be approximated by something very close to the actual observed distribution of outcomes. However, the variance has, in a sense, to be derived from a fitted model. This is one step further removed from reality which risks introducing more incorrect assumptions.
Other reasons are that the variance doesn't even exist for some assets (too heavy tailed distributions) and that errors in estimating the joint distribution can be smaller thanks to at least some level of central limit theorem.
Interesting analysis and I think I've reached similiar conclusions when talking to my friends and associates and seeing their results over the last ~5y; high risk leads to lot of capital loss often for them.
I usually have been playing low-risk investments and want to change that, any articles or advice to make independent high risk bets that actually end up lowering overall risk?
I accidentally wrote this comment backwards and I'm on a device that doesn't let me edit very well. You'll have to put up with some nonsense drivel until you get to the part you actually asked about.
----
I don't have much advice on that, specifically, other than
1. Be very careful. When shit hits the fan and you need that independence most, all correlations go to 1.
Related to the above, understand the cut-throat nature of the business. Actually, reading The Poker Face of Wall Street by Aaron Brown might help with this. The difficult part is not winning big; the difficult part comes after that, when you need to convert people's promises into money and successfully walk out with it.
Nobody is looking out for you. It's nobody's job to make sure you get what you are supposed to. I bring this up because I see people spend lots of time concocting advanced technical strategies but then forget the people skills required to get away with it.
2. Learn and play around with as much statistics as you can find the time to. I like reading actuarial textbooks because many of them work in insurance which means they do specifically that.
3. I'm a big fan of Aaron Brown for putting into simple words many of the most important concepts. Read Risk Management for Dummies (honestly!)
In particular, once you start making specific bets, you don't want to have stop losses that prevent you from losing too much money. That's a sure way to lose money. Instead, think like a scientist. Do hypothesis-based trading. "I think that X, because Y. If I see signal Z from the market, one of my assumptions are invalidated and I will reject my hypothesis and exit the trade."
4. Set aside a small amount of money to practise with. Try to invest in ways that expose you to specific spreads, like big vs small market cap narrowing, or service vs product sectors widening.
Try to make these actual bets you believe in. The point of this exercise is that you will, at a low cost, see how often you're wrong and go back to a plain constant fraction rebalanced portfolio, with a small but solid mix of high and low volatility assets.
It's really hard to do much better than that.
----
Edit: upon re-reading your question, I think I misinterpreted it. Maybe you're just asking about how to find a small but solid mix of high and low volatility assets to have in your constant fraction rebalanced portfolio.
One start is reading Thomas Cover's 1991 paper on the Universal Portfolio. In order to understand that, however, you might need to back up a bit more into the history of E log X optimisation. The book "Kelly Investment Criterion for Capital Growth" is a collection of historic and modern papers on E log X, including Cover's. It's a very good book. Though it takes a bit of working through results practically to understand.
----
That will help you find a decent mix of high and low volatility. Specific numbers are no big deal, though. I would recommend at least 50 % low volatility, but some people (like Taleb) suggest up to 90 %.
The mix of high volatility is also not that sensitive to specific numbers. To avoid having to do difficult optimisation, just split evenly between whatever assets you want in there.
To avoid correlations, just look for things that seem like they're not cointegrated. There's no point in optimising this too hard either, because as I started out saying, when you need uncorrelated returns, they will be correlated.
It's interesting to crawl through some of the entries from leaderboard people. It seems like most of their bets are neutral or slightly random, but then they'll be propped up with a lucky call on GME from January or some other single-stock pick.
For example, this poster apparently just spammed GME posts during the run-up and got lucky, but if you followed their other calls (REAL and AI) you'd be deep in the red. Their GME spam puts them in the top 3 medium traders with a 236% 1-month return, but everything else they've posted has performed terribly: https://duedilly.io/trader/r.Ottikarottiii/
It would almost be more helpful if there was an option to subtract out the well-known meme stocks like GME. Some of these leaderboard entries have not so great results for their own calls, but they happened to post about GME at just the right time to dominate their average.