We have a strong Quantitative research team within Morningstar globally. A lot of work has gone into the background to create and mimic the way analysts look at data and attributes analysts look at to rate them. This is done through machine learning and artificial intelligence. Quant ratings augment the analyst ratings by introducing MQR which also assigns ratings just like analyst ratings: Gold, Silver, Bronze, Neutral and Negative. This is done by a machine and not an analyst. The alphabet Q next to a rating denotes that is MQR.
The MQR text is available in Morningstar Direct which is a subscription-based software for financial advisers/wealth managers and institutions.
What is the methodology of MQR?
We have been assigning analyst ratings to funds in many countries for a long time and in India for the last ten years. We have a lot of expertise on the quant research side building machine learning models. This particular model tries to mimic the decision-making method of analysts. The model has studied the decisions made by analysts in the past and tries to identify through analysis which are the most important attributes that an analyst would look at in terms of data points for each of the pillars - people, process and parent. The other thing this model does is that there are two separate models created for each pillar – the probability of being positive or negative. There will be certain attributes that will point towards people pillar rating being positive. There will be certain attributes more tilted towards the people pillar rating being negative. So each of these models – positive and negative would run separately and will derive a probability of rating being positive or negative. It will then arrive at a common pillar rating for the people pillar. The same thing will happen for the Process Pillar. It is all done by a machine learning algorithm.
The process is slightly more nuanced for the Parent Pillar. If we already have analyst coverage on the Parent through some other strategy that we cover then that rating will apply to all strategies within the parent. But if it is not covered by our analysts already, then the same concept applies which I explained earlier. The probability of alpha generation net of fee is one of the attributes while assigning Gold, Silver or Bronze ratings. If there is a very limited probability of net of fee alpha generation, then it will be rated Neutral or Negative. The only difference here is that it is being done by a machine/algorithm rather than analyst.
How accurate are MQR ratings?
The efficacy of the algorithm ratings over various periods studied by us has been good. It needs analyst ratings to derive the data but once that is done the algorithm can actually go ahead and populate a lot of uncovered universe.
How can it be used by investors?
The caveat is that it is done by a machine that clearly comes out with the Q connotation in the rating. But it is important to note that the efficacy of MQR has been good. If you want to compare funds which are quantitively rated, you will notice that funds that have been rated higher have traditionally done better over the long term on a risk-adjusted basis. So that definitely is a proof of the pudding. If you wish to check a fund which is not tracked by our analysts, you can definitely use the Quant Rating to filter funds based on individual pillars. We have recently augmented MQR with smart text. In our analyst rating, there is a lot of commentary on what we like and dislike about a fund/strategy. For MQRs, it did not exist till now. Now we are able to generate text which explains the readers why a strategy is rated for what it is. For instance, if the People pillar is rated above average, there will be text explaining the attributes that are working in favour of the People pillar and what are not and similarly for other Pillars which are Process and Price.
