![]() ![]() Euan Sinclair pointed out (on one of his Tweets) to a book that his colleague recently published. When they are secretive with their own employees, do you think someone will let it out to strangers?ĭr. After all, the entire firm survives on the fact that they have working strategies. You need to see how secretive these people are. These days, I am looking for a quant job, and am working on an unpaid internship at a HFT firm. I even took a certification course on algorithmic trading where no real life strategies were taught. All I got was trolls commenting with nasty remarks. I used to ask the same question a while back on Reddit and other forums. Way easier to trade the second derivative of the markets than the first. ![]() Ie when is the market likely to have more upside than downside vol, etc. They are predicting when you have benign vs volatile market environments. One other thing - without giving away the special sauce, the best signals tend to be ones that aren’t predicting an asset price or market direction. My suggestion to you is to read white papers, understand the concepts, and just experiment. So in this way, you end up using signals that are working and cut signals that for whatever reason aren’t. The best guys are doing a few things - one, they are updating the algos using machine learning techniques as new data comes out, second, they are watching how the algos perform vs their statistically backtested expectations, and third they are diversifying across a bunch of signals/strategies and allocating more $ to winners and cutting losers. And models and signals may not update for this. For example, cross asset correlations may shift depending on Fed policy, or oil price as an input to stock prices can switch polarity (increasing prices are good and reflect demand growth at low price regimes, but are bad and reflect inflation at high price regimes). The other challenge is that many findings and signals change as relationships in the market change. Then it becomes a statistical game over a long period of time. For academic findings, you have to basically find ways to translate these into thresholds of when to buy and when to sell or get out, which takes market knowledge and experience. Plus, you can whittle away returns with transaction costs. So yes, signal may result in statistically significant higher returns over a long time horizon, but on a day to day basis, you can get smoked or hit stops. However, they are usually describing a conditional state when the XYZ variables are in play, and the results are a DISTRIBUTION of outcomes. For example, white papers may describe a market phenomena (variables/inputs XYZ result in higher than average future returns over time horizon T). The signals I have that work best usually are based on academic white papers, but there is skill required to parlay these into successful trading strategies. You’ll find examples for inspiration and code snippets showing how to actually run them, but rarely anything that actually works as-is. Frankly, good algos are proprietary and extremely valuable obviously, so nothing that good will be shared publicly.
0 Comments
Leave a Reply. |