Hey I have a data science background. This is good and entertaining fluff, but if you know what you're looking at, there are important things missing and some things that look good but don't make sense. I give him 4.8/5 for making it look important. For example: He has two time series (S&P500 and the bonds) and he compares them to many possible offsets. This is called cross-correlation analysis. It's a real thing, but it's also notorious for overfitting the data and showing spurious relationships if you misuse it like OP dies here. When you test many different offsets, you increase the probability of finding a high correlation *somewhere*, purely out of random chance. This is kind of like flipping a coin and getting heads 10 times in a row; it's impressive if you only flipped the coin 10 times, but much less exciting if you flipped it 10 million times. You were bound to get a 10-head steak at some point. An overfit predictor is one that performs very well on the historical data used to find it, but poorly on new, unseen data. If you select the single best lag based purely on the highest R-value from your historical test (precisely what OP did here), you risk overfitting to random noise that exists in the sample, but isn't truly predictive. And that's almost surely been done here and the validation should have been on showing that the model isn't overfit. To validate a model like that you wouldn't back-test (what OP does). Some things you could do are split the data into in and out of sample (e.g. make the model based on only the first X days in the series, and then judge it based on its ability to predict the data after day X). You should/could take steps to remove seasonality or trends within the time series first (which we already damn well know the stock market is seasonal, so him using untransformed values is most definitely increasing his calculated correlation). It would also be good to do bootstrapping to check statistical significance, instead of just p value. But it is very entertaining. OP probably also has a data background, to be knowing what to do to specifically torture the data this way.

---

> Besides, wtf do a bunch of children know about market sentiment? It ain't stock brokers and investors playing RuneScape. It's like seeing patterns in clouds and concluding they were formed with intent. A bunch of nonsense. It's not an impossible correlation - one could argue that maybe when consumers have more disposable money they are willing to drop more money on online games. It's just you'd likely find better correlations.

---

Yeah, this was one of those questions that doesn’t need an answer, unless I’m blatantly wrong then someone correct me. I forget the word for that kind of question lol

---

I was in a similar situation with TSLA a few years back. The point I eventually decided to sell at was when I could sell 2/3 and stop working, and keep 1/3 in play. And by the way, there are ways to hold SpaceX more directly if you're an accredited investor.

---

RS3 does not have an api or give out access for GE data, even the reflected price on the ge is nearly never accurate(+/-10%). I appreciate the weaponized autism, but you brought a knife to a gunfight and I have over a year of playtime. This is either an excellent shitpost, or a waste of time based off fake data from an AI. 10/10 would read again

---

You can do wheel on an index too. Sure, you can get burnt when a black swan event or a solid correction tanks spx or qqq by 20% and it puts you out of the game for few months, but at least youre not risking getting assigned on a shitco.

---

I have tried to buy puts on carvana like many others  . Most of the times I have been slaughtered  It’s one of the most manipulated(at low volume) stock . And most certainly WSB isn’t buying long now.  I do believe there will be an offering soon , likely next few weeks. They have 4.8B debt long term and $700M PIJ debt due in 2026. They have no choice but to dilute 

---

Can you make an ETF that rebalances based on this signal? 😹

---

The data they used was readily available online so I decided to attempt a replication of this to prove them wrong, but it seems like there may be some substance to this. I have used OSRS bonds rather than RS3 bonds since there is a kaggle csv available for download. I have used the multi-day log returns using look-behind windows, and computed ACF with equal sample counts per lag. This is with checking through 40 day lag: https://i.imgur.com/yDtATBh.png And this is checking just through 10 days: https://i.imgur.com/i2PI3SO.png And again through 10 days but using a 2 day log return window: https://i.imgur.com/m04HZYB.png I would encourage someone else to also attempt replicating this analysis to ensure no errors have been made. Edit: btw I am using an exogenous ACF methodology

---

For the picks an shovels play, buy $LUBE

---