Log in
Account
Account
Tools

Study · · By Shashank

Does ranking stocks differently in top sectors help?

Verdict

Tested, not pickedBacktest, hypothetical, costs included

Adding a second ordering measure lifted Birch to 29.8% a year from 27.2%, and to 26.5% from 24.1% without its 5 best winners, but random orderings came close.

On this page

The short answer

Would Birch do better if it ordered stocks inside the strongest sectors by two measures instead of one? It beat Birch in most checks, but not clearly enough to rule out luck. So we have not adopted it yet.

Birch first picks the market's strongest sectors, then chooses stocks inside them. We noticed that the order in which it chose those stocks barely mattered: a random order did about as well. A separate study had found a second measure that adds information beyond a stock's recent rise. So we tried ordering stocks inside the strongest sectors by both.

How did we test it?

  • The change: a different way of ordering stocks inside the strongest sectors, combining recent strength with that second measure. Everything else stayed the same.
  • Three other orders: the second measure alone, recent strength within a narrower group of stocks, and a random order repeated 200 times (the luck test).
  • Every result averaged over 6 start dates from 2016, with Zerodha delivery charges and three levels of slippage.
  • The hard checks:
    • re-run without the 5 and the 10 best winners;
    • 2023 onwards on its own;
    • the biggest 50 or 100 stocks removed;
    • nearby settings;
    • a paired bootstrap of the daily returns, which re-draws the days many times to see how much of the gain could be chance.

What did we find?

Not adopted yet. It came closer than anything else we have tested.

CheckBirchNew ordering
Growth per year27.2%29.8%
Without the 5 best winners24.1%26.5%
Without the 10 best winners20.5%22.3%
At the highest slippage25.5%27.2%
2023 onwards, without the 5 best winners14.8%19.3%
Worst fall27%26%
Trades per year4748
  • The new ordering beat Birch in all 6 start dates on growth. With the best winners removed it won in 5 or 6 of 6, at every cost level. It was ahead in 10 of 11 calendar years, and behind only in 2018.
  • Most nearby settings did as well or better. The weakest roughly tied Birch.

Three things keep it from adoption:

  • The luck test. Without the 5 best winners, it beat 89% of the 200 random orders. Our bar is 95%.
  • The bootstrap. The gain is about 3.7 points a year. But the 95% range, which holds the gain in 95 of every 100 re-draws, runs from −0.4 to +8.1 points, so it includes zero.
  • The biggest stocks. With the 100 biggest stocks removed, the advantage disappears: 28.2% a year against 30.3% for Birch.

What failed?

  • The second measure on its own lost to Birch without the best winners: 22.9% against 24.1%.
  • Narrowing the group first did the same, 22.8%, with a deeper fall of 30%.
  • What we learnt about Birch: a random order inside its sectors does about as well as a careful one. Its edge comes from choosing sectors and from its breakout positions. Which stocks it picks within a sector adds little.

What are the limits?

  • The price data holds only companies still listed today. That flatters every result here.
  • Birch's settings were chosen with 2023 onwards visible, and we tried four orders before this recheck. The new ordering is a candidate, not a finding.