High Conviction Beats Low

What We Do 

Every quarter, thousands of professional money managers publicly disclose exactly what they own and how much they own of it. Our system reads those filings and measures not just what institutions hold, but how much conviction they have — who is adding, who's exiting, and who's making outsized bets. We distill that behavioral data into a single score for each stock, ranked from strongest institutional conviction to weakest. In the midcap example above, over 49 quarters of live history, stocks in the top quintile have outperformed the index by +0.69% per quarter while bottom-quintile stocks have underperformed by −0.94%, producing a persistent and actionable spread of 62.5% annualized. Persistently through ups, downs, and Covid markets.

What is Conviction? 

It can mean many things. We have engineered eight features that measure institutional conviction — how aggressively managers are sizing positions relative to their own portfolio norms, whether high-conviction holders are entering or exiting, and whether that conviction is accelerating or fading. A composite of these conviction signals, produces approximately 6.5% annualized long-short spread above the index (in this case the IJH ETF that tracks the Midcap 400).

What is a Conviction Feature?

That is our secret sauce but one scrubbed example is what we call Ownership Conviction.   

This feature asks a simple question: among the professional managers who hold a given stock, are they overweight it relative to their own typical position — or is it just another name in the portfolio? Each manager can hold dozens of stocks, and each stock receives a specific portfolio weight — say 4.2% of the portfolio. But 4.2% means something very different for a manager whose typical position size is 1.5% versus a manager whose typical position is 5.0%. The first manager is making a statement of conviction; the second is underweight. How much the manager is over or underweight is a ratio that can be measured and, in aggregate, paints a picture of overall Conviction for that stock.

The Model

The Delaware101 model uses eight different features to build the conviction composite score - some are anti-signals that predict crowding, and some are two-quarter signals that demonstrates how conviction builds and fades. The combination of eight features is a single conviction score for each stock, each quarter, ranked from high to low. Quintile 1 is 80 stocks with the lowest scores, and Quintile 5 has 80 of the highest conviction scores. It is our first public model - robust and utilitarian - to demonstrate the power of the Conviction factor.

Each of our eight conviction signals captures a small but statistically significant edge — exactly what you'd expect in efficient markets where easy patterns get arbitraged away. Crucially, the signals are largely independent of one another, measuring different dimensions of institutional behavior: conviction levels, positioning momentum, and crowding risk. When combined, these uncorrelated edges stack, producing a composite score that has separated top-quintile winners from bottom-quintile laggards by approximately 1.6% per quarter over 48 quarters of index history. Drop a note and we'll send you the scores.