Aside from all the retail rumblings in this thread, I really love the fact that the US is pushing itself out of the crypto market more and more. Sure it's a big market but for once Asia is much bigger, so when US retail isn't allowed to trade this stuff anymore nobody will notice
A German Grayscale ramps up - VanEck Launches Bitcoin Exchange-Traded Note on Deutsche Boerse https://www.coindesk.com/vaneck-launches-bitcoin-exchange-traded-note-on-deutsche-boerse Blackrock's Chief Investment Officer: Cryptocurrency Is Here to Stay, Bitcoin Could Replace Gold https://news.bitcoin.com/blackrock-cryptocurrency-bitcoin-replace-gold/ Blackrock? Is that the same firm that manages Jerome Powell's money? https://wallstreetonparade.com/2020...n-corporate-bond-bailout-program-for-the-fed/ You have a choice, you can try to figure things out all by your little self or you can just follow the money? Bring on the regulation. Bring on PDT, KYC, AML, SEC, CFTC, it won't matter. Big money want's some Bitcoin candy and they are going to get it. The sell off is a gift for them. Volatility exists to transfer resources from the weak to the strong.
lol, and the concepts that will cut to the core - distributed and permissionless. The crypto industry will innovate. How are they ever gonna regulate an individuals vpn login to a DEX based in a crypto friendly jurisdiction? This sell off is a gift to those whom understand BTC’s value prop.
Fractional Differencing. I can generate single ticker weights (one time thing) pretty quickly on a 28 thread (14 core i9-9940X) workstation. If you're using a large universe (5000 stocks or so) you need to pass it off to a good GPU or you'll grow old watching it. Once you've generated your weights you want to test for stationarity with something like the Augmented Dickey-Fuller test The benefit of FD is it will pass the tests for stationarity and still allow for long and short term memory associated with financial time series. This can drastically help ML models scratch meaningful signal out of a low-signal-to-noise dataset. It's not perfect, you can't use a single FD feature and expect good results (multiple FD's tend to be highly correlated and not additive to the model) and if you transform the original FD output too many times you can end up with even less signal then you might with just standard integer differencing, log etc. i.e.: Don't calculate a rational-difference from the mean of a fractional-difference. The only thing you'll get from that are false positives. Like anything it's one tool (that you should have) in a well organized box. Additional background reading from: Hudson and Thanes Ritchie Ng Ritchie Ng's Github That should get you started and Google can toss you deep down the rabbits hole if you're inclined to go there. I also highly recommend de Prado's "Advances in Financial Machine Learning" to lay the ground work for any FML project. Chapt. 5 covers frac_diff as well.
But not BTC. Still 7-10 transactions per second? Luckily we are not using it as form of payment... For comparison, Visa does 1700 or so.
As you keep reminding everyone. Currently, it’s value is expressed as a base, a gateway, security and settlement - not payment throughput. Other blockchains and Layer 2 solutions are where those advancements are being developed.