Many of you probably know this, but for those who don't, historical data suggests that whenever US Dollar Index (DXY) goes up, BTC goes down and vice-versa.

This became the inspiration for a recent undergrad project that we did. We were tasked to make a model about anything and put it in an ACM-formatted paper so I decided to go with this, but it's been a month now and I don't think we'll be publishing it anytime soon since it's waaayyy too simple (and perhaps a bit flawed) imo. So instead, I will just share some of it with you all. <3

Methodology

The BTC dataset came from https://www.blockchain.com/charts/market-price and the DXY dataset came from https://www.kaggle.com/datasets/balabaskar/us-dollar-index-data. Yes, I know. The DXY dataset is a bit outdated, but we had to make the paper in a week (actually we were given at least 2 weeks but we procrastinated for the most part) and we couldn't get a better one in time.

Based on the BTC rainbow chart, we added a DXY price parameter to the model and fitted the model. I won't go into the details since it will probably bore you and I'm sure you're just here to see the results, so without further ado...

Results & Discussion

Here's the resulting fitted model: ^{(these are all rounded to the nearest 0.01})

BTC Price = 10 ^ (-17.71 + (-0.06)(DXY Price) + (3.30)(ln(Number of days since 3 January 2009)))

This model got RMSE = 0.25, R-Squared = 0.96, and Adjusted R-Squared = 0.96

Here are the graphs in comparison to real data:

​

​

Logarithmic Chart (yes, I know I could've scaled this better)

Linear Chart

You can kinda see that the model managed to capture many upward/downward movements (except 2019, idk what happened there) despite not getting the actual price.

Conclusion

Honestly, idk if this could help anyone since DXY isn't really that predictable either. But hey, it looks cool and we even got a good grade for it.

​

Inner Thoughts

^{Should I really be posting this here? I'm kinda worried about what would happen if all of a sudden they want to publish this sht. lol}

​

EDIT: I forgot to say, we did split the dataset for fitting and testing. The images there are the results of testing.