Share this page!

Enter URL or ID to Unroll

You can paste full URL like: https://x.com/threadreaderapp/status/1644127596119195649
or just the ID like: 1644127596119195649

How to get URL link on X (Twitter) App

  1. On the Twitter thread, click on or icon on the bottom
  2. Click again on or Share Via icon
  3. Click on Copy Link to Tweet
  4. Paste it above and click "Unroll Thread"!
  5. More info at Twitter Help
cephalopod Profile picture
cephalopod
@macrocephalopod
Zuu sufu sufu chyu chyu chyu tsu zuu fufufuuu...
53 subscribers
Jan 30 • 4 tweets • 2 min read
Thursday morning quant interview question. A junior comes to you with a ML model trained using walk-forward validation, and shows the following backtest, created by stitching the out of sample periods. What are your comments? What might they have done wrong, if anything? Image I think 4-5 people got this exactly right, and a few more had answers along the right lines but didn't mention some key detail. This is the first definitely correct answer that I saw -
Jun 27, 2024 • 9 tweets • 3 min read
Correlation between your signal and future returns is an important metric in quant trading. But what is a “good” correlation? Here’s a simple way to think about it. We’ll use a simple model where future returns y over some time period tau are normally distributed with a mean of beta * x and a daily volatility of sigma (here x is a signal with std deviation 1) Image
Jun 1, 2024 • 6 tweets • 2 min read
Interesting discussion, and follow the thread for further discussion of whether risk models cause crowding or not.

My view — they don’t really. Maybe a little on the margins, but the main drivers of crowding are alpha-driven rather than risk-driven. In quant firms, proprietary signal research can uncover new, idiosyncratic alphas (which causes firms to decorrelate). But over time these ideas diffuse (researchers and PMs move between firms and take ideas with them) which causes them to correlate and crowd into the same names.
May 29, 2024 • 6 tweets • 2 min read
Does the profitability of vol selling strategies depend on starting volatility level?

A short story. We start with front month VIX futures beginning in 2005, shortly after the contract was launched, so ~20 years of data.

For each day, calculate the P&L from shorting one futures contract. By working in price space we ignore any issues from from calculating VIX returns.
May 17, 2024 • 11 tweets • 2 min read
One more post about RenTech because it gives me an excuse to talk about Sharpe ratio, autocorrelation and scalability.

One thing that’s a bit surprising about RenTech returns is that their Sharpe ratio is “only” about 2.1 gross and 1.9 net. With volatility of 30% on gross returns and 20% on net returns, that translates into 60% returns gross and 40% net.

These are great results by anyone’s standard, especially sustaining it for 30 years.
May 16, 2024 • 7 tweets • 2 min read
Many mistakes here, including confusing gross and net returns, and not understanding the the fund mostly paid out profits as a dividend, so you couldn't compound.

So if you invested $10,000 into Medallion at the start of 1988, how would you *really* have done after 30 years? It's pretty easy to figure out, since the net returns are listed along with the fund size at the end of year year, so we can approximately know how much capital was allowed to remain within the fund and how much was returned.
Mar 7, 2024 • 13 tweets • 3 min read
A couple of people asked how to price this bet. As a reminder the bet Peter offered was 5-1 against that BTC/USD would hit $100,000 before the end of the year (i.e. he receives $20,000 if Bitcoin hits 100k, and he pays $100,000 if Bitcoin does not hit 100k) Intuitively that seems mispriced, but how can we sharpen that up a bit? Let's convert it to a derivative contract. The bet (from Peter's pov) is equivalent to paying $100,000 to buy a contract that pays out $120,000 if BTC/USD hits 100k.
Jan 14, 2024 • 11 tweets • 2 min read
I feel like some people talk about quantitative/systematic/automated trading as if they are all the same thing, which is not true, and blurring these lines causes confusion for people who want to enter the industry. “Automated” trading (contrasted with manual trading) is the simplest to understand. If the strategy doesn’t require any human input as part of its execution, then it’s automated. If there is a human in the loop then it’s not automated.
Jan 10, 2024 • 4 tweets • 1 min read
Something useful to keep in mind if you're using ridge regression. The normal equations are below, where the parameter lambda controls the amount of regularization. Image If you've normalized the columns of X so that the diagonal of X'*X are all 1s, then there's a nice interpretation of this - ridge regression both shrinks X'*X toward the identity matrix (ignoring correlations between features) and shrinks coefficients by a factor of 1+lambda Image
Jan 1, 2024 • 6 tweets • 2 min read
Here's a cute story about risk management and how we traded a name that ran up in price from 250 to over 500 in a couple of hours, and then back down to 200 in about twenty minutes. This is the price chart of $TRB over the last 10 days - Image We were long this name for a while (a little unusual - the holding period for this strategy is around a day) and here I've normalized our dollar position by dividing by our max pos. We won't buy any more when this ratio is above 1, and we start selling when it's above 1.15ish Image
Dec 6, 2023 • 7 tweets • 1 min read
A lot of quant workflows basically look like "do these somewhat related but basically different tasks and then combine the result into a normalized output"

For example - - Pulling together data from different sources, cleaning, mapping symbology etc
- Creating a large number of features as part of a fitting pipeline
- Fitting different models to get a forecast
- Searching over different parameters/strategies in a backtest
Jun 16, 2023 • 4 tweets • 1 min read
This actually seems kinda positive, especially compared to what people were speculating on at the time. bloomberg.com/news/articles/… The Chinese securities held were commercial paper of ICBC, China Construction Bank and Agricultural Bank of China. These are some of the biggest banks in the world, they have trillions of dollars of assets and decent credit ratings.
Jun 9, 2023 • 9 tweets • 4 min read
Quant interview question. Each dot is an average over a few thousand executed orders, x-axis is participation over the execution period (signed order qty divided by market qty traded) and y-axis is how much the market moved over the period. Why is there a discontinuity at 0? Image The quantities are all rescaled/renormalized btw, they don't tell you anything about the answer.
May 23, 2023 • 4 tweets • 1 min read
Addis Ababa Guadalajara
Apr 2, 2023 • 4 tweets • 1 min read
Nice alpha in this thread. Reminds me of the “up-modify” trick people use on some exchanges to drop lots of small orders in the book, then size them up when they get near top of book without losing queue priority. Not everywhere has this but iykyk. Another great trick is on exchanges that allow users to put stop loss orders directly in the matching engine. If you know what to look for you can identify these and it can be a strong signal that the price is going to move a *lot* once it hits a certain level.
Mar 19, 2023 • 4 tweets • 2 min read
We’ve known this for a while but the last couple of weeks made it clear — trying to analyse what a VC says in terms of true/false is a waste of time.

Applies to @jason @balajis @DavidSacks and many others. True/false is not a dimension that concerns them, nor is right/wrong or whether something is good/bad for the world.

Literally the only thing that matters, the only frame through which it makes sense to analyse what they say, is whether it pumps or harms their bags.
Mar 16, 2023 • 8 tweets • 2 min read
Apparently bookmarks are going to be public now? Anyway here’s a thread of all my bookmarks
Mar 9, 2023 • 11 tweets • 2 min read
I spent some time looking at FTMO and it is absolutely, 100% a scam.

They're not even trying to make money from profitable traders. They make money from people repeatedly paying to take the "FTMO challenge" in the hopes of being allowed to trade a real money account. How does it work? You pay a fee to take the "FTMO challenge" which is a 30 day period where you trade an FX account. The fee varies depending on account size, from $155 for a $10k account to over $1000 for a $200k account.
Feb 17, 2023 • 4 tweets • 1 min read
You’re telling me that a guy who sells online coaching courses is saying that getting a coach is the key to success? Oh my god it’s even more meta than I realised, he sells online courses which coach you to build your own online coaching course.

“It’s no use, Mr James …”
Dec 29, 2022 • 4 tweets • 1 min read
This is a weirdly common failure mode in smart people who are new to finance. I can see the attraction - if you know the distribution of some process, then it opens up a load of analytic tools you can use to 'solve' for the right thing to do - but in practice it never works. Real-world processes almost never have neat little distributions that are amenable to analytic solutions, so you save yourself a lot of trouble if instead of trying to find the "right" distribution, you just use tools/methods that are robust to the choice of distribution.
Dec 11, 2022 • 4 tweets • 1 min read
Day 9. Borrow availability.

A lot of factors have good alpha on the short side. In particular it’s pretty easy to find shitcos where you can be pretty sure the price will drop. To put the trade on you first need to secure borrow in order to go short. And if you try this you’ll find that you can’t actually get borrow for a lot of the most attractive opportunities (this is true in developed markets and doubly true in emerging markets)