Detail broker data & trading on information rather than noise

Detail broker data

We now have data feeds for the complete set of StarMine factor data and, far more importantly, detailed broker estimates and recommendations.

The broker data enables us to calculate our own consensus forecasts. These will include:

• Flash Consensus: This only incorporates broker forecasts that have been revised following what we define as a news event - for example, an earnings announcement, which almost always triggers changes to all broker forecasts. Analyst forecasts that pre-date the news event will not be included. Defining a news event for the purpose of Flash Consensus is more complex than it may seem. Our methodology is based on detailed analysis of historical events such as earnings surprises and profit alerts, and how brokers subsequently adjust their forecasts. The Flash Consensus will be used in our earnings revisions factors.

• Smart Consensus: This applies analyst weighting, time weighting, and revision weighting methodologies. There will be two versions of Smart Consensus forecasts. One will be used to construct a new sentiment factor, based on the difference between the Smart Consensus and the Market Consensus, normalized by the standard deviation of broker forecasts. The other will be used to calculate valuation factors derived from forecast data.

OQ is funding this data purchase. We do not - and will never - use a passthrough fee model.

Trading on information rather than noise

In December 2014, while working at Macquarie Bank, we moved to a more traditional systematic quant investment process. It didn’t work as was probably the biggest mistake of my career.

We underestimated the friction costs of trading - particularly market impact costs - and did not properly account for stocks that were not suited to our investment style. The biggest problem, however, was that too often we traded on noise rather than on information.

Consider what happens after a company announces a significant earnings surprise. The share price reacts immediately, but analysts take time to update their forecasts. The length of this lag varies considerably. Large, well-known brokers tend to move more quickly than smaller firms. In some cases, analysts even publish a report acknowledging that they will revise their forecast, but delay releasing the updated numbers until they have more information - for example, after attending a management briefing.

Unless the investment methodology properly accounts for this time lag, there is a significant risk of trading on noise rather than information. For example, after a major profit warning, the share price will fall immediately, but earnings forecasts may remain unchanged or not fully reflect the profit warning for some time. During this period, the stock can appear cheaper than it truly is. In other words, the valuation factors based on forecast data will be distorted, and trading on this basis is effectively trading on noise.

Our backtests did not account for this issue - and yet the results appeared very strong. As is often the case, the long portfolio trended upward while the short portfolio declined steadily toward the x-axis. At first glance, this might suggest the problem is insignificant. Of course, that conclusion would be false. Backtests are almost always flawed, and relying on them exclusively when developing an investment strategy is deeply misguided. I have developed a healthy skepticism toward backtest results. In particular, it is nearly impossible to avoid overfitting, especially when the objective - as is so often the case - is to maximise backtest performance statistics.

I believe there is no substitute for investment experience. From many years of trading, I know that one must be cautious about relying on quant signals after a significant news event. During this period, many signals are compromised. However, this window can be shortened by making smart adjustments to the data, and that is precisely what we are working toward with our Flash Consensus forecasts. We believe that incorporating detailed broker estimates alongside our own consensus forecasts - rather than depending solely on the market consensus - marks a major step forward in the evolution of our investment process. It is an exciting development.