Some price zones contain little historical volume, so once price enters them, there is less resting interest to slow the move. This creates a liquidity vacuum where momentum can accelerate until price reaches the next high-volume node.
The inputs are normalized returns and MACD-style momentum features. That keeps the experiment clean, but it also means the paper is mostly testing whether attention over historical trend regimes improves trend-following, not whether the model discovers a broad transferable market structure.
One of the strongest biases a data scientist might bring to the industry is the belief that they need hundreds of features to predict something, when in reality they only need one (but a useful one). That's where research begins and nonsense ends.
The trading signal you research is never isolated. It is always linked to a specific context. The same signal can be valuable in one context and useless in another. Researching a signal means researching the conditions under which that signal expresses an edge.
These nodes often act as temporary balance areas because buyers and sellers previously agreed there. Trade the first clean escape from the node only when price leaves with acceptance, not just a wick.
Although the paper has several obvious flaws, including look-ahead bias, survivorship bias, and an excessively short out-of-sample period, it still provides an interesting comparison of portfolio construction methods such as MVP, HRP, and HERC. However, the key issue is the results themselves, they are mediocre.
The biggest pitfall is that the paper treats higher backtest returns as evidence of model improvement. It stacks several DQN extensions, tests them on a very small number of assets, and then reports large returns without enough controls to separate real edge from overfitting.
Large deviations from VWAP are faded only when the tape slows, spreads tighten, and immediate momentum weakens. This avoids shorting strength or buying weakness while the move is still being driven. The trade targets reversion back toward VWAP, not a full trend reversal.
The authors admit that post-2004 EB predictions are less accurate because the 20-year rolling window fails to adapt to the market regime shift caused by information technology. EB could be interesting under a stable data-generating process, but markets are not stable
10K Followers 1K FollowingGestor +11 años Derivados+Vola+Spreads+Padre de Familia. Tweets no recomendaciones https://t.co/gOU6keGwP4 https://t.co/7KmW0wTJv5
3K Followers 310 FollowingHerramientas para tomar el control total de tus inversiones:
#AmplitudDeMercado, #Amibroker, #Sistemas, #BaseDatos, Alertas de #MarketTiming y mucho más.
10K Followers 337 FollowingAll views systematic: https://t.co/hz2IJAhW7R & https://t.co/mGbPf9PtMf. Quant of the Year by Risk Magazine: https://t.co/7MeiDASJc2
61K Followers 1K FollowingCIO at https://t.co/eTgt1vPj4z
Winner of the 2023 World Cup Championship of Futures Trading. Awarded as Best Audited Trader 2023/24 by Rankia.
4K Followers 3K FollowingTrader pro +20 años con track record público. Opero cada día con sistemas reales. Ahora enseñando trading algorítmico para que ganes con datos, no con emoción.
75 Followers 897 FollowingNetwork Engineer.
Não fui eu que lhe ordenei? Seja forte e corajoso! Não se apavore, nem se desanime, pois o Senhor, o seu Deus, estará contigo. Js1:9
25 Followers 814 FollowingSpecializing in Mercedes-Benz body part new and Grade A, fairly used. Quality component to keep your Mercedes looking and performing at its best. DM for inquiry
14K Followers 134 Following10+ års erfarenhet. Delar insikter & analyser för att hjälpa dig att utvecklas som investerare.
Bluesky: https://t.co/Xlt6VTwmCk