Smart Alpha is now live

Dear All,

The redesign of our model exchange platform is now live. We made some changes from your feedback.

Another interface change that was released was a menu re-organization under Portfolio. We’re now grouping mainly by whether a system is live or simulated. The “New” menu was also simplified to only show New commands in context. For example, when viewing simulations, there are only two options now: New Stock Simulation and New ETF simulation (before there were 6 options, very confusing).

Let us know what you think.

Thank You

Great Job Marco and team,

I like the changes. Everything I need is still available and it’s easier to use.

Thank you,
Mark V.

Marco,

Thank you for continuing to improve the sims without limiting them.

Better to make changes as to what is presented on the Smart Alpha page–without limiting the sims themselves–as you have done.

Well done!

Jim

Do I have to be a Portfolio123 Subscriber to follow Smart Alpha models?
“Free Portfolio123 Subscribers are only allowed to subscribe to paid Smart Alpha models.”

Marco, who are the free P123 Subscribers?
I don’t see any free memberships being offered under “Pricing”.

Geov, We’re working on an overall redesign by year end. This was the first step.

Georg - I assume what is meant is that people who subscribe directly to a model but didn’t pay for P123 subscription won’t get the freebees.

Steve

On the ‘My Smart Alpha Launched Models’ page, I can no longer see how many members are “watching” each model. Will the members still be able to follow the same models? Or will they need to re-follow on the updated site?

Like it!

Chris - the watching count should be back - thanks.

Member subscriptions have not changed.

ted

We no longer have the ability to see the model behavior over the full 1999 up to launch period (in-sample period), and then see its behavior since launch (out of sample). what has happened? why can’t we see the full simulation results over the maximum data available like before?

Marco,

It would be nice to see the comparison to the benchmark for volatility and MDD on the risk score presentation.
(Why the risk score using only the 90-days period and not since launch figures?)

I think the “Risk score” should be as a default on the Multisort and also it is important to add an “Annualized Return Since Launch” filter as a default (maybe instead of one of the return or excess return figures - we don’t need them both).

Please add the amount of exposure and/or the average weight for a single position near (or instead of) the number of current holding presentation or turnover.
(I think it is far more important than both current of holding and the turnover figures)

Regards,

Ohad

Thanks Ted.

Marco, would you be so kind to disclose more about

Of special interest are you going to disclose model design over-optimization risk in any way?

Is it possible to download model daily performance (see no button near the chart)?

As an investor subscriber in light of recent developments I’ve already checked how can I cancel my membership.

There is a Download button, under the performance chart, bottom left. This gives daily performance in excel. You can then analyze this sheet for whatever you want.

Georg, thank you. So this is the way to check model for over-optimization…

P123 Log and Download buttons are not visible on small resolution screens (attached).


download.jpg

We thought our new “rolling tests” would help in detecting possible over-optimization. The main idea was that, by moving the start date and running a bunch of three year simulations, the curve-fitted models would NOT show consistency. It didn’t work as expected. Models that fell apart after launch were not identified by rolling tests. They fell apart because they were not robust to changing economic cycles, psychologies, etc, things that the past 16y of data does not have.

So we’re thinking of other ways to help investors choose a set of strategies.

First thing that had to go was the backtest results. This shifts the focus of a designer from stressing out about past results (to compete with other models) to future results.

Next , we have introduced ways to search for particular Style, Themes, Caps, etc, to achieve better diversification. Hopefully designers will add models where needed. For example there isn’t a single model that focuses on the highly uncorrelated ‘Special’ Theme (Energy, Precious Metals, Airlines). We’ll soon publish a paper on Themes and their correlation with each other.

We’re also going to add a blog feature where designers, and anyone on p123, can write articles and gain respect & followers. The idea there is ‘know your designer’. P123 is simply a tool to automate a strategy, consistently and unemotionally. But creating the strategy is a painstaking , careful process, that requires experience, and should come from someone you trust.

There’s more to come. Ability to auto-trade, run combinations, easier interface, etc.

Hope you stick around.

Marco,
thanks for the update. Highly appreciated. This is going into the right direction. Emphasis on OOS is good.
Still waiting for European Data.
And: I will SURELY stick around!
Werner

This is a great point by Marco: “They fell apart because they were not robust to changing economic cycles, psychologies, etc, things that the past 16y of data does not have.”

Sub optimal out of sample system performance is not just due to over optimization. Successful systems need to both consider what did happen and what could happen as the past rhymes but never exactly repeats. Furthermore investors in general need to be more patient as every model can and will have periods of under performance that are impossible to detect prospectively. It is also helpful for investors to reset their expectations from making 50%/year (impossible to obtain consistently with any significant amount of capital) to beating their benchmarks (achievable after costs)

The rolling test tool is an excellent addition to assess for performance stability in lower turnover systems and I am glad that we have it

I do not want to digress from the core topic any more than I have so please continue on with the smart alpha posts.

Scott

Marco - what are the rules for model revision now? Is it still every six months?
Steve

We can’t – but Mr. Market and Father Time can and will do so loudly and boldly.

This has been hotly debated in the forums pretty much since R2G was launched almost three years ago, and there is still no consensus answer (as Marco noted, we hoed rolling tests would be revealing but it did not turn out that way).

I think the problem is caused by the fact that the things about which we complain really have nothing to do with what steps a designer took or did’t take or even whether a model is “robust” or not. It has to do with intention. Is somebody intending to design a model that succeeded in a specific and closed environment (the sample period)? Or is somebody designing a model with the intent to perform in the unknowable future?

If one intends to design something that succeeded in a specifically defined and closed-ended sample, sound quant processes can be and have often been shown to have been very successful. So, too, can design based on investment principles.

If one intends to design something that has a reasonable probability of succeeding more often than not in the unknowable future, then exclusive reliance on quant processes cannot work; one can get lucky if and for as long as the future more or less replicates key aspects of the sample period. But over time and as the world evolves, that sort of thing cannot be sustained.

There are four ways to measure designer intent:

One is to reveal the trading systems, but for obvious reasons, that is an absolute no-go.

Two is to consider designer-written descriptions. This can be as effective or ineffective as designers choose to make it.

Three is to try to objectively measure stylistic choices, as we have done with the style ratings (actually, these don’t measure designer intent; they measure what the designer achieved whether or not it was intended).

Four is to rely on the team of Mr. Market and Father Time. The good part is that as time passes, this is foolproof. The market will do what it will do and whoever does what Mr. Market wants will be rewarded and we’ll objectively know how well models stand. The bad part is that it may require patience for new models. But that’s the way the world is. New asset managers, mutual funds, ETFs, quant boutiques, etc. have to tough ti out waiting to accumulate live track records. Newbies in every walk of life need time to show what they can do. Our current platform acknowledges that we have to live in this world; we can’t change the world to get outcomes as quickly as we wish.

Our old sim-oriented presentations kept the over optimization, curve fitting etc. conversations alive far longer than should have been the case. As time passes, and we live with the new approach, I believe we’ll see that this line of conversation will diminish and eventually vanish from Smart Alpha related discussions. (It will likely persist, but in other threads relating to how one should approach strategy design, which is where that topic should be discussed).