Replicating “Models, Regimes and Trend Following - Part 4” A White paper by. @JungleRockRes . Get your copy at Part 4 colab link:. Code available and ready to run (including data) on Google Colab for the complete series at:.

vibe coded a yield curve sandbox just for fun.check it out:.

RT @internQuant: Replicating “Models, Regimes and Trend Following - Part 4” A White paper by. @JungleRockRes . Get your copy at https://t.c….

As always, I'd like to thank the whole @JungleRockRes team for these papers. I have learned so much doing this series I could not be thankful enough 🙏. Also, thanks to everyone that has tagged along and checked the series out. Feel free to reach out if you have any questions,.

Finally, we select a handful of strategies and plot their cumulative return:. - Most of the strategies look fairly similar to the ones shown in the original white paper, I noticed the ones that combine TSM signals tend to be a bit worse in general in my replication, i could not

Now we extract signals from our forecasts, build strategies accordingly and use TSM strategies as benchmarks to compare our MSM modelling to. - As discussed above, the fitted MSMs in this replication tend to classify bear regimes more accurately, which generally leads to

After obtaining all of our forecasts, we compare our predictions to a model that has been fit on the full sample of our data. Similarly to my part 3 replication, my fits seem a bit better at classifying regimes(specially bear ones). I can only assume this is due to different

Then we run the forecasting process to obtain estimates in between model fits, as well as training Random Forest models to predict probabilities using the lagged filtered probabilites and moving averages of returns and squared returns. This process is also time consuming, but.

Next, we jump into the forecasting framework that fits a new model at the start of each month. We end up with a total of 296 fitted MSMs, I implemented a neat workflow to run these fits so it runs a bit faster, this process took around 10 hours to complete in my machine, more.

As always, left is my replication, right is screenshots from the white paper. Part 4 starts by demonstrating 2 approaches for forecasting MSMs:. - The first one is a forecast approach by not feeding in new information(actual returns), using forecasts to forecast further down

tradings bonds over IB feels like trading items in an MMO lol.

RT @JungleRockRes: Would you allocate to the .Jungle Rock ETF at 69 bps?. It would be a multi strategy using trend following, asset allocat….

RT @Templarpilled:

intern mobile just hit 4k Km 😎. been with me since the beginning of my internship

RT @JungleRockCap: Like this post and share . 2 giveaways to the research portal of .Jungle Rock, our sister company . Retail Tier.

finally starting to get some encouraging results. seems the msm implementation from python’s statsmodels is a bit better at classifying bear regimes, ends up changing some of the paths/combinations. Looking pretty similar to me tho, what do you think?.

👀

RT @JungleRockRes: My top 15 memecoins, now like it & share 😎

After fitting 296 MSM models, one for each month in the sample, Part 4 holds onto each model for the length of its respective month, using and updating its estimated parameters each day, obtaining new probabilities and inferring regimes. These 3 methods do exactly the month-wise

RT @JungleRockRes: like and repost to play the jungle rock raffle.