Hacker News: mathrando

New comment by mathrando in "Tether reserves backed by 2.9% cash"

mathrando — Fri, 14 May 2021 07:25:45 +0000

> They are creating a dollar substitute and basically running a banking and payments business but without the oversight that anyone else doing a similar kind of business would have.

You could take that word-for-word from the debate that raged after Treasury bailed out the prime funds in 2008.

New comment by mathrando in "Tether reserves backed by 2.9% cash"

mathrando — Fri, 14 May 2021 07:08:41 +0000

I expect better from FT. Tether is very similar to prime money market funds, in both structure and portfolio composition. I guess the Fidelity's of the world aren't keen on competition.

https://www.nytimes.com/2008/09/20/business/20moneys.html

https://www.nytimes.com/2020/03/19/business/coronavirus-mone...

New comment by mathrando in "Show HN: I built uFincs – a privacy-first, encrypted personal finance app"

mathrando — Wed, 12 May 2021 05:38:42 +0000

Why not team up with the ProtonMail people to build a browser extension that verifies and logs javascript sigs/hashes? Corporate clients may like it. Gives them an IOC for the next big supply chain issue.

I don't know enough about browsers or js to know if its difficult or not.

New comment by mathrando in "A review of correlations between big five personality types and life outcomes"

mathrando — Sun, 09 May 2021 18:38:36 +0000

Articles like this answer the question, "what if 1920s eugenicists got hold of 1980s magazine relationship tests?"

The scary thing is a lot of commercial "people analytics" systems marketed to HR departments, lenders, and government agencies are little more than dressed up relationship tests from 1980s magazines.

https://www.cs.princeton.edu/~arvindn/talks/MIT-STS-AI-snake...

New comment by mathrando in "The Time Everyone “Corrected” the World’s Smartest Woman (2015)"

mathrando — Sun, 09 May 2021 02:55:16 +0000

Whether Monte Hall is counter intuitive is a function of how the question is phrased.

When you phrase it in a way that underlines the mechanical nature of the host's decision, people get it right. When you phrase it in a way that suggests the host's choice is itself random, people get it wrong.

I think the first formulation primes people to think of it from the perspective of the host, which is the right perspective for this problem.

New comment by mathrando in "Advancing Excel as a programming language [audio]"

mathrando — Thu, 06 May 2021 01:34:41 +0000

> Probably the worst example I've seen first-hand was an entire retail banking loan-approval process

There's some real doozies out there. The Reinhart-Rogoff error, which was used to justify imposing austerity on Greece [0]. The UK's COVID tracing fiasco [1]. The list goes on. People using Excel have no business making decisions that affect other people's lives.

[0] https://stanfordreview.org/clarifying-the-implications-of-th...

[1] https://www.bbc.com/news/uk-54422505

New comment by mathrando in "LSD, cargo shorts and the fall of a tech CEO"

mathrando — Wed, 05 May 2021 18:44:24 +0000

you're doing it wrong. first you drink all the booze. then you hack all the things. if your investors aren't down with that, they're in the wrong line of work. maybe they should consider haberdashery.

New comment by mathrando in "The art of solving problems with Monte Carlo simulations"

mathrando — Tue, 04 May 2021 12:56:15 +0000

You still need to integrate over x. y|x=c is Gaussian for each c, but there are many such c's in [a,inf].

If you're curious:

https://www.ams.org/journals/mcom/1978-32-141/S0025-5718-197...

New comment by mathrando in "The art of solving problems with Monte Carlo simulations"

mathrando — Tue, 04 May 2021 11:36:35 +0000

RNG quality is probably the least of your worries.

The curse of dimensionality means you often need a huge number of samples to guarantee a useful level of precision. Additionally, many quantities you'd like to know come up in Monte Carlo simulations themselves. That creates an intractable O(N^2) problem.

Here's an example that I encountered at work yesterday:

You want to know P(x>=a and y>=b) where (x,y) is a standard bivariate normal with correlation R.

You can:

1. Simulate a bunch of (x,y) and take the mean of the indicator function 1[x>=a, y>=b]. This will take whole seconds to compute.

2. Use an adaptive quadrature method like scipy.integrate.nquad. This will take dozens to hundreds of milliseconds.

3. Write a fixed grid Gauss-Legendre quadrature routine in c. This will take less than a microsecond.

I needed this value inside a Monte Carlo simulation. If I had to simulate the solution, the problem would be intractable.