If You Are Not Paying You Are the Product

I am really starting to think something has changed. Not only are the stocks not working as well but the price action is different.

I will note here that some excellent Designers are doing best in the micro, micro-cap models that even include OTC stocks where institutions may not play.

While the market makes new highs some of our models are suffering. Mine are. The Designer Models could be doing better. There is one thing that has truly changed as far as data.

Robinhood sells its trading data. Not just the order flow but the data. It is a type of alternative data that Renaissance Technologies, D.E. Shaw & Co., AQR Capital Management and all the rest of the Quant Hedge Funds now have access to.

I do know that Renaissance Technologies has found that newsreaders can be used as a signal. Supposedly the number of times a stock is mentioned—without need to know if the comments were good or bad—is a useful signal. No doubt they have looked at social media where “You Are the Product.” Can you imagine what you could do if you had real-time insight to what people are actually doing with their money?

There is no reason to think RobinHood is the only one. A Forbes article that supports what I am saying: HERE

From the article:

“Another means of making money without charging the customer is to sell order-flow DATA. This can disadvantage you too, because the data can be used by other market participants to arbitrage activity around short-term trades, potentially your trades.” Emphasis mine, of course.

Anyway, some of the Quant Funds named above may know what you will be doing today. Maybe before you do. I do mean in general terms—to the extend that the people at Robinhood are trading like us. And again, I am not sure that it is just Robinhood.

They will have a tougher time moving ETFs with their $$TRILLIONS. $2 Trillion is a conservative floor just adding the assets of the big names. This is not counting leverage. If things do not change I will be moving to ETFs. I am still hopeful. I have not moved out of my ports—not completely anyway.

-Jim

IB seems to be doing it a lot less than others - though I am not clear why they would somehow do it “a little”. They say they are not focusing on it which is different than saying they are not doing it. They show some revenue from it but a lot less than the others (assuming the chart is right)

https://www.interactivebrokers.com/en/index.php?f=44074#section-pfof-drop

Either way, if your systems are working less well, is it because you do not get the price you thought you would get?
If your fills vs expected are an issue then order flow selling might be the problem indeed. If not, it is more that some factors (e.g value) are temporarily out of favor though I have seen value picking up significantly in my systems in the last couple of months.

For my part, I keep things simple and place limit orders on Monday using the Friday close +/- a tiny %. I see no difference in fills in the last couple of years vs prior. I use only large caps.

Jerome

Thank you Jerome for your personal experience with this.

Be aware that payment for order-flow that you link to and selling the data of what retail investors are doing (why not real-time if you are going to do it?) are two different things.

I do not like either and if you do one you might be doing the other so your comment is pertinent no matter how you slice it. But the two are different.

Appreciated.

-Jim

Could be also simply a normal underperformance streak of you modells.
I you have a value tilt in them, value is doing terrible right now, also momentum modells where killed in 2018 and 2019.
Also we had a lot of quad4s (inflation slowing, Gdp slowing) and quad3s (inflation going up, gdp going down, that is happening right now), that
favour industries stocks and low vola. (see attachment).

Quad3 (happens right now!) for example favours energy and kills small cap value.

My tips:

  1. try to add industry momentum to your modells and see what happens, helped my ones a ton!
  2. Momentum works in three (1-3) of the four quads (market regimes), value only in Quad1
  3. Quality works in three quads (2-3)
  4. be out of stocks in quad4 (inflation slowing and gdp slowing) for timing

This is also the reason why those ETF modells are pretty consistent, bc. their exposure is clear on a lo of different themes like beta and defensive asset classes (bonds etc.).

The quad stuff is from Dalio and Hedgeye (i find their calls hard to follow, but theire framework helped me with my modells)

Also:

Even very, very good modells do not always track the market, I even find that robust ones (assumtion!) tend to undertrack regualary (going up, when markets going down, or going down, flat when markets going up).

Best Regards
Andreas


Andreas,

I agree this is credible explanation.

Not that I will ever be sure that Robinhood has my interest at heart. For sure, someone thinks that information is valuable enough to pay Robinhood and help them write their regular salary checks.

I wish to add that there is a lot of opinion that we should have known which ports (Designer Models) to pick. You provide, here, a rational approach for rotating types of models (or adjusting the allocation).

I am probably coming to this from an entirely different direction. I am also looking at methods of allocation also, however. Maybe Georg addresses allocation and/or market timing too.

Thanks for the ideas. Good ideas, and frankly I hope value comes back in addition to possibly learning some allocation strategies.

Best,

-Jim

Some of my thoughts on the market and the diluted impact of somewhat effective Value factors historically correlated to returns:

[quote]
Jim: While the market makes new highs some of our models are suffering. Mine are. The Designer Models could be doing better.
[/quote] There has been an obvious disconnect between the market and many quant models for much of the time at least since the beginning of 2017. I believe that the market might be returning to some “quant normalcy” in some respects, as noted below.

From 1/2/2017 to 9/2/2019 the “Basic: Value” ranking system, applied weekly to the S&P500 equivalent universe with Percentile NAs Neutral, showed strongly inverted results. From 9/2/2019 through yesterday 11/20/2019 the results are not inverted and are what we would normally hope to see, that Value factors have some importance. Back in early June I noted the inversion and the implication that the most expensive stocks had been strongly favored. Ranking based on just market cap, higher is better, shows a similar strong preference for large cap stocks until 9/2/2019.

The beginning of September might be an inflection date, when the odd (to me) market behavior ceased. Reasons? Some of you must have a better list than me. My personal belief is that the market reacted naturally to forces that P123 data and perhaps FRED data do not reflect. Movement of large amounts of money, perhaps seeking perceived safety into U.S. large caps? I would hate to think there has been market manipulation, but it is possible. The market as a whole has enjoyed very good returns through the suspect period and my returns have been decent, so I am not complaining, just trying to understand the why of it.

To sum my personal position up: I am in the market with the models I created (“dancing with the one who brung me”)!

The simple explanation is that the abnormal markets were a result of Fed Reserve aggression with interest rates and an end to Q.E. both of which (initiatives) unsurprisingly failed. We are now back on the path to zero interest rates and likely more Q.E. The life we are starting to see in smallcap value stocks is probably a result of the return to easy money.

SteveA

At worst we’re at a top, at best a consolidation phase. It’s also coming into holiday season when things get dull. In a few weeks managers will be locking in their 2019 performance, so I wouldn’t expect much action until January.

Jim, add the following ranking and see if your trading systems improve…

<Composite Name="Industry Momentum" Weight="22.2%" RankType="Higher">
	<IndFactor Weight="25%" RankType="Higher">
		<Factor>Pr4W%ChgInd</Factor>
	</IndFactor>
	<IndFactor Weight="25%" RankType="Higher">
		<Factor>Pr13W%ChgInd</Factor>
	</IndFactor>
	<IndFactor Weight="25%" RankType="Higher">
		<Factor>Pr26W%ChgInd</Factor>
	</IndFactor>
	<IndFactor Weight="25%" RankType="Higher">
		<Factor>Pr52W%ChgInd</Factor>
	</IndFactor>
</Composite>
<Composite Name="Quality (Gross Margin / ROE)" Weight="11.1%" RankType="Higher">
	<StockFactor Weight="20%" RankType="Higher" Scope="Universe">
		<Factor>GMgn%TTM</Factor>
	</StockFactor>
	<StockFactor Weight="20%" RankType="Higher" Scope="Universe">
		<Factor>GMgn%Q</Factor>
	</StockFactor>
	<StockFactor Weight="20%" RankType="Higher" Scope="Universe">
		<Factor>ROE%TTM</Factor>
	</StockFactor>
	<StockFormula Weight="20%" RankType="Higher" Name="ROE Ratio TTM/5Y" Description="" Scope="Universe">
		<Formula>ROE%TTM/ROE%5YAvg</Formula>
	</StockFormula>
	<StockFormula Weight="20%" RankType="Higher" Name="GMgn Ratio Q/TTM" Description="" Scope="Universe">
		<Formula>GMgn%Q/GMgn%TTM</Formula>
	</StockFormula>
</Composite>
<Composite Name="Accumulation/Distribution (InstitutionalSentiment)" Weight="11.1%" RankType="Higher">
	<StockFormula Weight="50%" RankType="Higher" Name="Inst Own Ratio MRQ/PQ" Description="" Scope="Universe">
		<Formula>Inst%Own/Inst%OwnPQ</Formula>
	</StockFormula>
	<StockFormula Weight="50%" RankType="Higher" Name="Vol Ratio 1Mo/3Mo" Description="" Scope="Universe">
		<Formula>AvgVol(20)/AvgVol(60)</Formula>
	</StockFormula>
</Composite>

Andreas,

This is certainly a good idea and I will look at the out-of-sample results. Too bad I, and many of the Designers, were not smart enough to do this a year ago. Too bad I cannot go back and change history. But backtests are a beautiful thing that might work going forward.

SteveA and SteveT,

Very plausible theories. What evidence (going forward) should I look at? As I understand you are expecting value models to do better going forward?

If what you say is true will the designer models be beating their benchmarks at a year. At 2 years?

Should I just look for the Designer models with a high value rating and see how these are doing at a year (or maybe 2 years)?

Generally, what evidence can I look at (going forward) that will give me some information as to how credible these theories are?

Personally, if both my (present) models and the Designer models are underperforming in a year, I will be changing my ideas of what is the most likely explanation.

For now, I hope you are right and I think there is a good chance that you are. But I have been predicting a turnaround myself and the turnaround theory gets a little less credible every day for me.

Maybe it is just the trade war—another credible theory I think. I am not saying I know.

I will say I am pretty convinced that something changed but that it could be cyclical. Going forward–until I get more evidence–I will be looking at more conservative (and rationally based) allocations and possibly cyclical adjustments of the weights I put into ports/ETFs. I do not believe that this is inconsistent with what your are saying.

In fact, SteveT, I think I may have gotten my allocation strategy from you: IE., the Tangency Portfolio or maybe the Minimum Volatility Portfolio. Maybe use MPT to find the optimal portfolio using predictive models for the expected returns.

This last would, in essence, be a cyclical adjustment of the allocation: of the port/ETF weights.

-Jim

Andreas (or others),

Do you have more details on this Quad approach (links etc)?
e.g. I do not understand why you state that Value only works in Quad1. On the picture Quad1 states “Eq style factor Overweight = Hi Beta, Mom, Qual, Growth” - but not Value…

I remember the same picture with possibly more details was posted years ago on the forum (I think it was already by yourself Andreas)

Thank you

Jerome

Hi Jerome, I don’t want to speak for Andreas, but just throw out what I can share on this. I’m pretty sure that image is from Hedgeye and is part of their rate of change process. (I followed them for a while after Andreas’s suggestion and learned from it.) Eric Basmajian on seeking alpha I think has a process that uses similar rate-of-change analysis for for looking at macro and I think his articles are very good.

I did a bit of math on it over last 20 years as well as since 1968, and if I did my numbers right (not a given as I’m not sure I’m using “right” rate of change in all cases), while returns coming out of quads 3/4 have been bad over the past 20 yrs, returns subsequent to quad 4 (slowing gdp growth and falling inflation) weren’t as bad over long term - perhaps due to sometimes getting big rallies coming out of cyclical bottoms, and perhaps due to more rapid historic cyclicality than we see presently. (I’m not sure about this. The latter is just speculation, just musing out loud.) Subsequent returns to the quad 4s in the 80s and 90s seem to do pretty well - such that quad 4 seemed like a signal to get positioned into equities over that time frame. Figuring out and positioning for the coming “quad” is a big part of what hedgeye seems to do though.

Thank you Michael / Spaceman.

While I do appreciate your answer, I am not completely sure what to make of it.

It looks to me that these 4 quads are broadly similar to the kind of economic cycles one can find in the literature - but nicely summarizing together the “better / worst” asset class, the “better / worst” sector etc for each Quad.

As you imply (I think), the key is to know in which Quad you are at time T and therefore it is a question of the indicators used and how reliable they are.

If I got this right, Quad4 is the typical recession / depression and indeed if you have a high degree of confidence you are near the bottom, it makes sense to start going heavy in equities. But I do not get the feeling from your backtest results that you do have such a high degree of confidence in the indicator. Or did I misunderstand?

Thank you,

Jerome

Hi Jerome,

ist not on the Picture, your are Right, I pickt it up from there Videos.
Explanation: When Inflation goes down and GDP goes up at the same time Cyclical Rally, e.g. a lot of value stuff…

My take: I would Have been dead in the water the last years, if I would not Have combined value with Momentum, value has ist
Merits, but it is not for me to wait years until it is in favour again…

Regards

Andreas

Hi SpacemanJones,

that is really, really interesting, I did not backtest it by myself, just some real time experience with hedgeye calls.

If somebody knows how to implement this in a trading System (GDP / Inflation rate of Change and its combinations that would be great.

Am I Right, that it is possible to include buy and sell Signals out of a macro Chart fom P123?

Thank you

Regards

Andreas

I have recently been researching the hedgeye quads. I was able to create the historical quads using data from FRED for inflation and GDP and Excel for backtesting. It’s basically looking at the change in the growth rate for inflation and GDP. I compared three years with hedgeye and only one quarter was different, but I think that is because the data changed.

What hedgeye does is predict the current and next 2 quads. And they have a playbook based on the quads for sectors and style. But they also apparently have a large analysts team that provides additional validation before making investment calls. I don’t have a lot of experience, but I like their approach and subscribed. I also recently subscribed to EPB Macro. Its playbook is based on changes in the growth rate, so one dimension instead of two.

My assumption is that since hedgeye predicts the future quad, that it’s possible to backtest using the historical quads. On a go-forward basis, the newly predicted hedgeye quad would then be updated manually in the formula for the current quarter.

Yuval described in a post a strategy for ranking sectors based on historical performance such that those sectors that outperform for a given ranking system are ranked higher, as one of many factors. My thought was that perhaps this could be enhanced by overlaying the quads such that the sector ranking may change depending on the quad. Yuval also described a substitute for sectors called clusters in a post that might be used instead.

Using the quads, I assume other factors could also be weighted differently such as value. Currently, I don’t have any performance results.

Jerome and Andreas, I did this a long while back, but I’ve attached a spreadsheet that shows the calcs and what I was doing. I updated recent quarters so data should be fairly fresh. Columns bg and bh are 2 variations of quad calcs I used, and down at the bottom of the columns are the column sum and average forward return calcs for the quads. Hope it may be useful to you. Beyond just being familiar with the economic cycle, I don’t really utilize any of this for timing. It did lead me to a likely different view of quad 4 already mentioned.

Again, I’m not sure of the proper way to calc because I wasn’t always matching quad determination (and sometimes quads change with data change), but I think the calc variations used are getting the spirit. I make no effort to predict future quad - just looking at things like "if current or most recent quad is X, what is forward 3m, 6m, and 9m returns in columns out to the right.

There’s other stuff in the spreadsheet that I don’t recall much of what I was doing, looks like making rank percentiles of growth and or inflation, but I’m not sure of my thinking here other than growth used to be a lot stronger than it is currently, so maybe slowing down from 5% growth isn’t as big of a deal of slowing down from 2% growth. But again - I’m not sure what I was doing.

The worksheet initially had a lot of links to other spreadsheet tabs. I’ve tried to remove all of those links so everything will calc, but if I missed something let me know and I’ll try to fix it.


quad_p123.xls (838 KB)

I have made public a custom formula called $quad_2000_2019 which will return the quad for a given date between 1/1/2000 and Q3 2019. The values for the quads were calculated in Excel; all the formula does is check the date range and return the precalculated quad number. The formula actually calls another set of formulas, because of P123 formula length limitations.

SpacemanJones and I have the same calculation for GDP but different calculations for Inflation, so sometimes I have a quad 1 where he has a 2 or vice versa, and sometimes he has a 3 where I have a 4, or vice versa. I used GDPC1 from FRED for GDP as did SpacemanJones. I used CPIAUCSL from FRED for Inflation using the “Aggregation Method: Sum” on FRED to convert from monthly to quarterly whereas I believe SpacemanJones may have used PCE directly from BEA for Inflation.

Hi Greg, thanks for the posts. If you don’t mind me asking, have you used the quad information in p123 models?

I ultimately moved on to a different type of macro tracking, but have not utilized macro in my p123 models. I presently have the p123 models pick stocks, and if I need to adjust allocations for defensiveness it happens as an external decision. I know this isn’t ideal, but it’s what feels right to me and something I can more easily understand.

Thank you all - it will take me some time to review comments, formulas and spreadsheets and try to see if it can help me - but I wanted to thank you now.

My first test will be to use Gregg’s formula in an ETF system 1999-2019 and use for each each Quads a handful of ETFs representing the “right” styles & sectors (per hedgeye). Then compare that to SPY and R2000
NB: I would avoid doing pure binary shifts between ETFs but rather overweight / underweight. I have learnt that market timing systems perform better when viewed as helpful but imperfect rather than perfect 0 or 1.

Cheers