Methods for Combining Financial Data with Different Frequencies
Summary
The document surveys ways to combine fundamental, market, analyst, and alternative data in quantitative investment analysis despite their different observation frequencies. One option is to resample inputs to a common frequency; another is to train separate models for each data type and combine their forecasts in a meta-model. It also points to MIDAS methods, which are designed to incorporate predictors observed at different frequencies.
Other suggestions include building a signal from each source at the intended trading or rebalancing frequency, translating those signals into alphas and combining them, or modeling returns as a linear combination of features. Factor models illustrate how accounting information and price-derived momentum can coexist, while principal component analysis can reduce a large feature set to fewer dimensions. The document offers a menu of approaches rather than implementation guidance or comparative evidence. It does not address data-release timing, missing observations, leakage, or how to choose among methods for a particular strategy.
Key ideas
- Resampling can align data sources with different observation frequencies.
- Separate models for each data type can feed a higher-level model that combines their forecasts.
- MIDAS methods are designed to use predictors recorded at different frequencies.
- Signals from distinct sources can be tuned to a common trading horizon and combined as alphas.
- Factor models and principal component analysis offer additional ways to combine or compress features.
Tags
Full text
# Combining Quantitative data with fundamental data # Combining Quantitative data with fundamental data These day, there is relatively new phenomena of combining quantitative data and fundamental data called 'Quantamentals'. In this regards, I was wondering how to combine Four Essential Types of Financial Data 1) Fundamental Data(eg. Asset, Liability,sales) 2) Market Data(Volume, Price/Yield,Volatility) 3) Analytics Data(Analyst Recommendations, Credit Rating) and 4) Alternative Date(Satellite, Twitter) into one data frame for analyses, given difference in frequency. ## Answer by Azam Yahya (score 2) https://quant.stackexchange.com/a/39062 There are at least two ways of doing it: 1) Resampling them to their median frequency. 2) Build one ML model for each data type, then combine the 4 different forecasts into a single meta-ML model. (Courtesy: MARCOS LO´PEZ DE PRADO) ## Answer by pmichaels (score 0) https://quant.stackexchange.com/a/39090 Consider investigating the MIDAS approach of incorporating signals with different frequencies. The classical apporach is to create a signal based on each source and tune it to your trading/rebalancing frequency. Convert this signal to an alpha based on Grinold and Khan, then add the alphas together. ## Answer by Jacques Joubert (score 0) https://quant.stackexchange.com/a/44873 Keep in mind that if we turn to Arbitrage Pricing Theory then we can model the returns of the asset by some linear combination of the features. Take the Carhart four-factor model as an example it incorporates financial statement features with momentum which is a feature derived from closing prices. There are two books that provide a lot of context: Quantitative Equity Portfolio Management & the bible of factor investing Active Portfolio Management. Another common technique is to make use of a dimensionality reduction algorithm such as PCA to decompose all your different features into a few features that are information rich.
Shown in full with attribution under the source's licence. Licence: CC BY-SA 4.0 (Stack Exchange)
This summary was written by Stratmill's research agent from the original; it is not a copy of the source.