Performance measured against benchmarks
The ZZAlpha® machine
learning engine has been applied to over 20 market segments,
and results evaluated against benchmarks. As the tables
below show, in many instances over 5 year, 3 year and
single year periods, the ZZAlpha® recommendations
substantially exceeded their benchmarks. The results
come from recommendations that were certified each morning
before market open, to assure they support stringent audit.
We show here results for portfolios of 10, 5 and 2 equities, recommended daily (i.e. 50, 25 and 10 recommendations a week for each market segment).
We show here results for portfolios of 10, 5 and 2 equities, recommended daily (i.e. 50, 25 and 10 recommendations a week for each market segment).
Annualized returns for ZZAlpha® recommendation portfolios.
In 2010, we tuned the original algorithm based on the history
2006 to 2010 (which included the great recession). In
early 2021, we evaluated the algorithm with variations of the
parameter, and found there would have been better results for
the past 5 years with slightly different parameters in some
cases.
We have installed the new parameters in the production system
and hope to find that they perform better over the coming 5
years.Find returns for Q1 2021 here.
Find returns for Q2 2021 here.
Find returns for Q3 2021 here.
Find returns for Q1 2022 here.
Find returns for Q2 2022 here.
Find returns for Q3 2022 here.
"Results 2021"
"Results 2022"
Notes to tables:
1) Returns are annualized returns for multi-year periods, NOT average annual returns. Quarter, YTD and single year are returns for those periods, not annualized.
2) Benchmarks include dividend reinvestment but portfolios omit dividends.
3) We never use leverage or margin or options in evaluating results.
4) Due to a change in Sector definitions and memberships in 2018, we do not have sufficient data for some Sectors before that date.
Assumptions in performance
metrics
The results shown here are
produced by a mechanistic trading results simulator used to
evaluate all the ZZAlpha® machine learning engine
recommendations. (The evaluation is entirely distinct
from the machine learning technique.) The results shown
assume no trading commissions, no spread, no slippage and no
dividends. These results assume purchase at a price
equal to the reported opening price on the day of
recommendation and sale at the reported opening price five
trading days later. (Of course, traders often can and do
improve on this pricing assumption by scrutinizing the
market flow and news at the open or during the day.)
Real market testing
In testing by ZZAlpha
by actual market trading of various portfolio
recommendations for various periods, the actual net returns
track the evaluator model estimates.
Returns of professional
traders
The returns shown assume
mechanistic purchase at a price equal to the opening price
on the day of recommendation and sale at a price equal to
the opening price 5 trading days later. We expect that
professional traders watching market movement at open can
improve on these returns, or that speculators choosing to
sell in a 3 to 20 trading day window after purchase can
often sell above the price at open that we assume for
closing the position.
Better returns vs larger
scale
In general, smaller
portfolios have higher (but more volatile) results than
larger portfolios. Please contact us for
detailed information and statistics on portfolios of
interest.
* To be conservative, we often use both the selected index and equal weighted population means as benchmarks in our analysis of returns.
* To be conservative, we often use both the selected index and equal weighted population means as benchmarks in our analysis of returns.