1) Type 1: Understanding market structure and market incentives, and the corresponding flows
Type 2: Finding statistical anomalies within price/non-price data
Type 3: Hybrid approach.
Let’s seek to understand this further.
2) Type 1: There are many reasons for actionable price flows, such as price insensitive liquidation, factor premia et cetera. For an example, a previous thread
3) These have intuitive explanations. Assume an option writing strategy, specifically short puts. This should be a positive expected return strategy, since you are acting as the “fire insurer” for a market burn down. You take on the burden of risk and collect “risk” premiums.
4) Type 2: Finding statistical anomalies. Techniques in this domain are but many, including statistical and artificial intelligence methods. However, these are often abused by engineers, data scientists and “domain experts” who understand little about the markets themselves.
5) Their techniques, while ingenious, are often little suit for practical, financial trading. In GENERAL, increasingly inductive (2, 3, 1) approaches are likely to be more persistent, and have the added benefit of interpretability/conviction in its inference.
6) However, these approaches tend to be symbiotic, and a hybrid approach reflexively improves the process. That is, understanding factors driving the market guides the search for statistical anomalies, while statistical anomalies hints at relationships of underlying variables.
📌 Adopting Alpha Research Methods
7)
Type 1: Unfortunately, there are no shortcuts in understanding driving market factors. Reading macroeconomic analysis, understanding market players, seasonal effects, fundamental analysis are all helpful, but there is no one-size fit all.
8) A good approach is to read more literature, as well as reliable analysts who opine on financial events. In addition, financial literature and mathematics are also important.
9) It is reasonable that one trading options should understand the Greeks, as well as the assumptions behind the pricing model (BSM).
10) An accessible class of alpha that does not require understanding of “complex” macroeconomics or financial mathematics are Risk Premias, and a booked recommended by @therobotjames:
11) Book: Expected Returns: An Investor's Guide to Harvesting Market Rewards
These are often accessible to retail traders to harvest and relatively easy to intuit.
12) Type 2: 1)Yet again unfortunately, there are no shortcuts in understanding the theoretical underpinnings of statistical methods.
13) Standard AI algorithms often fail to find persistent statistical anomalies in a problem setting where the hypothesis space is both large, dynamic and probabilistic with low signal to noise ratio.
14) Alphas exist as sparse, non-persistent, decaying subspaces within an infinite-dimensional Hilbert space. Even worse, they have high tendency towards false positives.
15) Applying standard AI techniques almost invariably leads to absurdly high Sharpe ratios from alphas that overfit complex, non-linear unstable relationships between said variables, but are practically useless in predictive power due to the properties of the problem setting.
16) However, such statistical learning methods can become useful when “hacked”, often relying on heuristics designed to cleverly overcome said complexities that both reduce search spaces and counter false positives.
17) Design of such “algorithm-hacking methods” requires understanding the mathematics behind the algorithms, but also knowledge of a wide range of algorithms to be used under different settings.
18) The “hacking” part is then adjusted to the knowledge of the trading domain, which can only be achieved by yet again understanding (Type 1).
1\13 One of the biggest challenges in quant and alpha research is obtaining clean, error-free data. Models need be built, using assumptions to reduce “dimensions” of reality for tractability.
A THREAD ▼ ▼ ▼
Machine Learning, Sparse Datasets and Error-Free Simulations
2\ Mathematical models, by definition are built to simulate and capture some phenomena, practical or abstract. Often, they are built on data, which themselves are derived from some unknown, statistical distribution.
3\ For example, an alpha model, is backtested upon data where prices/returns are drawn from some distribution. Widely in quantitative finance, they are assumed to be drawn from a log-normal distribution.
Been a fulfilling ~1 month since our launch. For hitting 1k followers, we have a special thread for you, including a premium alpha report and a case study. 🔥
MEGA Thread (N = 60+) : Robust Alpha Research Processes; HangukQuant
1) We adopt the Hybrid (Type 3) approach in the alpha research process. The hybridity is reflected in our team’s dynamic, with practitioners working on the theoretical models, and traders providing input on the heuristical discovery of alpha.
2) The result is a coherent product in the form an “alpha report”, that premium subscribers get access to weekly.
1) Type 1: To keep my knowledge of finance, I both subscribe to financial literature, academic or otherwise. That means reading books/papers on finance, trading, economics, podcasts for general knowledge, and a working knowledge “Mathematical Finance”.
2)
Type 2: My knowledge on Computer Science and Statistical methods comes from my background in academia. I also keep up to date on new state-of-the-art research by reading academic literature.
1) First, what is alpha research? It is the undertaking of research operations towards finding excess returns on the market. It may come in many forms, from both fundamental and quantitative perspectives.
2) In the quant sphere, it often takes on the form of parametric formulae/rules designed to express underlying trading edge. They also come in many forms, such as stat arb, risk arb, factor models (mom, rev), idiosyncratic premia, predictive modelling et cetera.