EXDC is the world’s first annual conference bringing together leading minds in finance, business, media and sustainability to discuss the real-world use cases for alternative data.
The video could have divulged the returns and risks generated by leading ML based quant funds over the years and also returns generated by some major trading algos over the years, to bring out theshelf life of specific algos and ability to generate sustained higher sharp ratios.
I don't believe in hedging. Why not just lessen the position size if you're not that confident. On some occasions, both the original position and hedge can go against you.
Not believing in hedging is like not believing in clouds. It's true that in some occasions position and hedge can go against you; that doesn't invalidate the fact that good hedging strategies will lower the probability of losing by diversifying exposure.
De Prado's logic: since poor models fit badly I don't use any model. Unfortunately this approach is not going to work when data are not enough to tell you anything close to the truth, e.g. in finance where returns are fat tailed and most observations are just noise
@@albertosantangelo6872 You said that you don't have enough data and some return distribution has fat tails. And you believe your data is noise. You failed to grasp Marcos' message to maximize recall instead of precision. Recall can be converted to precision by the ensemble model. That's machine learning fundamentals, you statistical genius.
The high frequency example of flash crash is classic fat tail. When many algorithm act the same way you get extreme events. Without the common models, their actions would cancel. But if many players come to same conclusion, you get the black swan. Machine learning is causing the extreme events, the black swans, as it correlates activities.
the problem is that we lack financial market data, especially bearish market data. If the company focuses on HFT, it surely has enough data to deal with regime change. However, for medium-term trading, ML cannot help us to predict regime change because of lacking data.
Not sure what you mean by not enough data. We have nearly 15 years of decent T1 data for all asset classes - a lot of it is free too! Data Quality tends to increase with the year it was collected (aka more gaps, missing data from 2000’s than 2019) Besides, increasing your training set doesn’t necessarily reduce OOS variance.
Omega Sigma is right. We can easily create synthetic data sets that is seeded from historical data to fill in gaps or simply create more similar data. Just read the scikit-learn documentation on it
Keep up the good work. I really think that you deserve more views! Did you ever look into using smzeus . c o m?? You should use it, it will help you get the views that you deserve!!!