Risk: Thinking creatively

Manage
33

Quad - that is a very interesting paper.

"Utilizing the current capabilities of P123, the only tool available to use this information is market timing, by selling equities and going to cash when systematic risk is elevated. "

I believe that a book can be constructed that maximizes diversification without the use of market timing. This can be done by an appropriate mix of ETFs and stocks. However, market timing could in theory provide higher levels of return, albeit with higher risk. It really comes down to investor expectations. Some are quite happy to make 8% return per year provided it is reasonably low risk. Others aren’t happy with 50% return per year. If you are in the second camp then of course market timing is the only option as diversification won’t get you there.

Steve

34

I would agree that utilizing ETFs to diversify into other asset classes can help, but P123 is not necessarily the best tool (currently) to arrive at the strategic asset allocation. There are better (though not perfect) tools, such as mean-variance optimization models or Monte Carlo simulation, based on forward-looking capital market assumptions. That being said, I believe that allocation models are best used as directional indications only, since they are flawed (e.g., assuming normal distribution of returns, rebalancing every period), and often very sensitive to even slight changes in assumptions. I would never suggest “goal seeking” (i.e., optimizing) your way to an allocation.

How would you propose implementing an ETF portfolio? The ETF strategies that I have viewed on P123 seem to rotate into the “best” handful of ETFs (e.g., 1-2 ETFs at a time) based on historical performance (e.g., returns, volatility). How is this not market timing?

35

I’ve wasted way too much time on P123 threads in the past… so not wading in much, but the ‘diversification’ problem is fairly ‘simple’ at the book level. Invest in very conceptually different things/return drivers that have logical reasons for maintaining inverse/zero correlations in peak stress market environments. So, managed commodity futures and long-short trend following systems and short only stocks or short-term hi-quality gov’t bonds tend to exhibit this with long-only equity.

Can also hold ‘cash’ and, if you are willing to invest A LOT of time in learning, some dedicated ‘short only’ portions of the portfolio.

So, a simple port ‘book’ of a) personal income (from a job), b) cash, c) short-term, hi quality government bonds from the strongest economy nations, d) long only equity systems (can vary market cap, style, country, etc), e) managed futures or trend following long-short systems (long vol profile for the best managers in historical peak stress times), f) short term real estate or asset backed loans with margin of safety (short term senior credit investments, with collateral and say 60% LTV ratio), g) small’ish (or tactically varied) allocation to short only equity systems, and h) if you want, some market timing stuff… or (optional) i) selling vol through options selling after big vol spikes, with 2-3 year periods (or exit based on VIX levels) of investment followed by cash. There are some good market neutral traders out there as well, they can be added. Or rent generating real estate investments. It’s fairly good in many markets and offers inflation hedge as well. For pure returns, ‘angel investing’ is very good - if you begin at the bottom of an up cycle (we are now likely at the top).

The combination will give you a balanced overall portfolio that is very unlikely to all go down at once.

You can also build ‘simple’ binary systems - Bonds/cash, long-equity/cash, real-estate/cash, gold/cash etc. and allocate to some of these.

Many of these can be done in P123 with ETF’s and stocks if you invest the time, or can hire pro managers. Some can’t be done here no matter what.

A full discussion of ‘risk’ of any one system is something I’ve tackled in many threads in the past, so not gonna dive back in. But good luck.

As far as what is and isn’t ‘market timing’, it’s a giant ‘boogey man’ on this site. Any portfolio that differs from the market portfolio is ‘market timing’ in some way. Everyone market times. They just either do it intentionally or unconsiously.

36

Quad - would you agree that the traditional risk-based strategy of stocks/bonds/cash allocation based on age is a form of risk control? If so, then that is my answer. Yes it can be done. It can be done with stocks, bonds and cash, as a minimum. These can be put into ports without market timing and the book can do the % allocation. With the variety of options we have today I suggest it is possible to improve on that.

And if superior results depend on market timing, then so what? If the market timing can be done in a rational way (such as minimal correlation across securities) while adding more performance than risk, then isn’t that something that should be pursued? I think Marc’s mandate is to improve on what we have, not give up because we don’t already have it.

Steve

37

Tom - I agree with what you say but my position is that if there are tools that will help with the book-level diversification effort then why not have those tools available? I think you can make an educated guess about what to put into a book for minimum risk but you should be able to reduce the guess-work with good diversification tools. It is like driving a car. It may appear to be running well but you don’t know for sure without plugging in a diagnostics computer to tell you what is going on under the hood.

Steve

38

Changing the topic and thinking out loud a bit. We should have an “acid test” for company-level risk. Whatever Marc decides upon should be able to identify risk in various companies that have collapsed. A good example is FXCM , instantly put on life-support by government policy reversal. FXCM deals in FOREX which is high leverage, but is there anything on the books that would indicate such risk?

FXCM is also a good response for anyone that believes Black Swans are rare events.

Steve

39

Steve,

There are many things that can be done at the book level and I’ve posted on many of them in the past… there doesn’t seem to be any want or desire for them.

For example:
a) Allow people to create ‘rules driven’ books to auto adjust the total amount of cash invested in a book with mechanical rules - for example, this will allow risk targeting or beta targeting.
b) Allow developers to adjust the amount of hedging on a book based on rules - so, if there is a DD of 10-20%, can add X% hedge at the book level, if DD of 20% to 40% can adjust hedge amount, etc.
c) Allow developers to use buy and sell rules in a book with constraints, so we can set maximum position sector or asset class weights, or trailing period max correlation constraints, etc.
d) Allow for creation of long-short books, with ‘beta matching’ and total sector controls. This would allow professional-light type long-short systems.
e) Allow for user inputs on ‘total book level’ constraints - for example, cap size, long, short, position level, industry, etc.

I doubt many users would use these, but this is how professionals begin to use rules to drive books. As it is, if you don’t build the systems and someone else does, and subsystems are optimized, then combining them in books just makes it worse.

Best,
Tom

40

I also suggested year(s) ago that P123 save all launched R2G’s. Over time, can build real data driven models of what drives consistency over time in out of sample vs. in sample. Can look at things like total factors in ranking, number of hedges, total gross exposure, sector concentration, number of positions, etc. over several hundred systems from many designers. If no R2G’s are deleted, it’s valuable and can begin to extract ‘probabilities’ of certain factors driving consistency out of sample. However, this wasn’t followed, I think. And no need it should be, it’s not my company.

Something like Omega Ratio can be added (although it won’t detect overfit systems, it’s another useful - or Marc would say, useless ratio). I have seen studies of ‘ranking traders’ for 3,6 and 12 months go forward performance based on trailing period returns. Sharpe, Sortino and Omega all have some statistically significant positive results - although many multi-manager allocators claim to have much better proprietary models.

As far as ‘company level risks’… can identify the largest down day in a company’s history, as well as the largest down day (or 5 down days) in the past 20 years in the most closely correlated companies (on major factors such as size, financial quality, industry and ‘earnings misses / beats’ to this one). Can also look at ‘factor extraction’ to try to figure out which industry or company this company is most like on those factors and then look at the largest historical DD’s of the those most closely matched companies. Looking to create a multi-factor model that ‘predicts’ extreme events and ‘risk of ruin’ for this ‘type of company’ based on the companies ‘grouping’, not just the company’s history. Sorry, I don’t have more time to write more clearly now.

41

“There are many things that can be done at the book level and I’ve posted on many of them in the past…”

Tom - I believe that everyone would use books if they were much smarter.

Steve

42

" I have seen studies of ‘ranking traders’ for 3,6 and 12 months go forward performance based on trailing period returns."

I believe for mutual funds, this strategy can be useful if you are betting on the fund manager. As soon as the fund gets a new manager, all bets are off. (Take Templeton for example.) Likewise, I believe the same can be said for R2G models. It is probably more useful to bet on the model designer, not the model itself. Same goes for “ranking traders”.

Steve

43

Okay. I decided to ‘play’ and added many factors to the various google doc list. They are all brainstorm factors.

44

My earlier comments regarding diversification were regarding an equity portfolio of individual securities, which I view as P123’s core competency. It would be nice if P123 expanded its capabilities to better evaluate multi-asset class portfolios, but I didn’t think that was within scope of the original post. As I mentioned earlier, I use other tools for broad asset allocation, and P123 for equity security selection. I found evaluating ETFs on P123 to be difficult because there’s no information on fundamental factors (e.g., valuation, yield).

Regarding books, I like Tom’s suggestion to include book-level constraints on sectors, market caps, and beta. Improved support for long-short books would also be helpful.

1 Like
45

Using the 80/20 rule, perhaps 3 uncorrelated portfolio’s would reduce risk by 64%. Perhaps echoing what Tom said, in my experience, nothing reduces risk more than diversification among uncorrelated assets/strategies. The problem is to find good uncorrelated strategies to diversify into. Long stock portfolio’s will always have a fair amount of correlation. To truly reduce risk, I think, requires some different approaches.

46

Wow, lots of interesting things appeared in the Google doc! I have some questions . . .

Item: Presence of large positions as a % of daily trading volume by activist investors taking large positions in the stock and looking to ‘agitate’ for change

Question: While there is risk of decline if the agitator goes away, I suspect that most of the time, this is seen as an attraction, not a source of potential drawdown. In my experience, when there’s smoke, fire, if it isn’; there on day one, tends to follow eventually

Item: Senior Management issues - age of CEO (i.e. Buffett) or key players

Question: BRK is BRK, one in who knows how many. More often than not, however, this can be seen as a bullish situation, a special situation. It may spark a sale of the company, a major restructuring, or even simply fresh innovation from new blood. I think many investors would prefer to load up on situations like this if they feel hard times are coming, even to the point of paying a “death premium” in valuation.

Item: Industry wide Margin erosion; for example due to industry big dog trying to buy market share and put people out of business (i.e. the Amazon effect) - companies may be less stable when a giant market player is in their space

Reaction: May need to change that from margin erosion to ROE erosion. Margins are over-rated by journalists. If narrowing margins is accompanied by greater turnover, then the end result may be net positive (look at the history of technology)

Item: Collapse of a major industry ‘player’

Reaction: Could be bullish. Look, for example, at how the fall from the summit for the likes of AOL or MSFT played out. One giant gets hammered; losts of other companies flourish

Items: Earnings and/or Sales of Beta/Correlation to GDP and Earnings and/or Sales of Beta/Correlation to other companies in portfolio and other discussions involving, one way or another, correlation

Reaction: I’d be careful about doing too much with correlation. These things tend to be unstable over time and may bear some of the problems I expressed with share price histories; statistical report cards. In the economy and in businesses, correlations in general seem to be rising in many respects (likely relating to globalization) and I’m not sure new norms have yet been established. Notice that commodities, for example, are no longer widely seen as uncorrelated hedges against stocks. I would urge that no correlations be considered unless a non-quantiative explanation can be made for what we can expect of future correlations. Relying on purely statistical correlations could actually result in a dramatic increase in risk of disaster (besides leverage, I believe this was one of the LTCM problems and also was the bomb that blew up a lot of other hedge funds in the late 2000s).

Items: Sensitivity to Fama French Factors and Sensitivity to Barra (or similar) equity factors

Reaction: Maybe I’ve had too long a day; I’m not sure what these are getting at.

47

There were posts several months back with the introduction of the MACRO screen and its future ability to generate a ‘Market Score’. I think this would be an excellent GAME CHANGING addition. Having the ability to generate and perhaps even laminate several of these market scores together as part of a buy and/or sell rule and/or hedge rule would give much more flexibility. So I’m officially cheering for Marco MACRO tool. Say that three times fast!

I’d also like to second what others have said about Live Books. Utilizing some additional rules and perhaps even the MACRO tool to determine individual portfolio allocations would go a long way to mitigating risk based on market conditions. And I’d like to go into detail on one of those. The Dollar!!!

Everyone seems to still be focusing on interest rates and the forgone conclusion that the FED will raise rates sometime late summer or early fall. I believe the uber-strong US Dollar is a much larger issue. Not only is the fed out of ‘dry powder’, but the delay on rising rates has set off a feedback loop that the FED will have a hard time controlling. Rising rates will only strengthen the dollar. As we know, a strong dollar is a headwind for exports. The large-cap stocks that have done so well will find it ever hard to compete for market share abroad. Conversely China, Japan and Eurozone are embarking on their own QE which will further weaken their currencies. US Treasuries are looking ever more attractive to foreign investors as our bonds have positive (greater than zero) yields relative to new bonds issued recently in Germany with negative coupon rates. The strong dollar and weak Euro only exacerbate this process. Short of Yellen announcing QE 4, there’s no end in sight to the dollar’s dominance. A true disaster could be a Greece default as it will prompt a panic to escape the euro… and where do you think that wealth will go?

Marc, thanks for posting that white paper on Buffett. I, like most, am fascinated by WB. But he’s not really an investor in the traditional sense. And I therefore really have no hope of learning from most of the steps he takes. He’s got way more tools in his toolbox that virtually every other investor on the planet, especially the individual investor. Warren doesn’t have to worry about liquidly to slippage - He just buys entire companies; engulfing them into BRK.B! He’s private equity + activist + [the most interesting man alive] all rolled into one. Really, he’s a cuddly, grandpa-esque version of Gordon Gecko.

48

“I would urge that no correlations be considered unless a non-quantiative explanation can be made for what we can expect of future correlations.”

I don’t agree with this statement. Increased correlation means higher risk. There are fewer and fewer places one can use to diversify one’s investments. It isn’t a question of looking for a new norm.

Steve

49

:slight_smile: Actually, from what I understand about the negotiations that led to him not rescuing LTCM, he’s not all that cuddly when there’s big money on the line. But yes, he does have an arsenal about which we can only fantasize.

50

Yes, increased correlation absolutely does mean higher risk, much higher risk. I’d love to be able to own of truly uncorrelated or weakly correlated assets. My point, however, is that we can’t trust the data available to us to tell us whether or not we really and truly have a correlation problem because the data is unstable over time. That’s especially troubling now, given that correlations in general tend to be rising, and there is the Wal street adage that in times of crisis, correlations go to 1.00.

Hence the point of this exercise is to bypass risk-control solutions that depend on mathematics and statistics, which however sound they seem in simulation with historic data still have incredible propensities to fail going forward when real money is on the line, and to instead try to get at underlying characteristics that can be expected to make future correlations, etc. what they are.

Consider, as an example, that Company A decides to use cash flow to pay down its debt, while Company B, an industry peer identical in every way except that it wants to pump its stock price so it takes on a lot more debt to buy back shares. The only data available to quants may suggest A and B are highly correlated and ought not be in the same portfolio, but actually, in the future, the correlation may be much lower because of a significant reduction in fixed interest costs and hence eps volatility, and they may well belong in the same portfolio.

That contrived example assumed identicality in all respects except for balance sheet. Actually, though, inter-company variations are much more wide ranging, and companies and businesses evolve. That’s why I’d rather manage risk through fundamentals and not through isolated share-price statistics.

Again, my position has nothing to do with what we want. I’m urging a focus on what we can credibly expect to actually get.

Now if we can find some sort of sensible credible correlation assumptions to work with, by all means we should go for it. But I’d need a lot more than a data crunch to make that work for me. I’d need a good, sound, logical argument that could be advanced even by one who never saw any historical data.

51

USELESS JUMBLED POST TO WASTE TIME.

The purest, most logically sound ‘risk modifier’ is holding some cash and some shorter term, hi-quality bonds, and having a reliable income. The bonds / income provide for ‘stable income’ in flat markets and some minor protection in big down markets. The cash provides for security and the ability to buy stocks on pullbacks and rebalance.

Then you hold some min. number of stocks to try to protect against ‘being wrong.’ Some people think it’s 10 stocks, some think it’s 10,000. That’s we have markets.

But, when system wide liquidity vanishes and buyers disappear, or forced selling happens - the quality of a company becomes irrelevant in the ‘short term’ (and the short term can last decades), and will stay irrelevant until buyers flood back into the market. This can take days, months or… depending on the cause - forever.

So… market wide order book liquidity and sector and industry and stock level order book liquidity would be great, but P123 doesn’t have it, and is not likely to get it, right?

So, in many respects there is, basically nothing that can be done with a 100% long only equity investment to limit losses in ‘all market conditions’. That’s why many of us like working at the book level and would love better ‘book’ rules and functionality.

We are now in fairly ‘rare’ / unusual historic market conditions, and have been since 2008.

So, using only ‘fundamentally derived’ rules on company fundamentals vs. ‘true value’, etc - won’t help much if the market ‘panics’ and ‘stays panicked’ or market fundamentals change significantly (nuclear bomb detonations in major cities around the US), or whatever - you are losing a LOT, if not all.

And the markets have changed. Huge, historic amounts of stock trading is now being done by High frequency and computers with much shorter hold times. Initial studies show that these conditions raise future expected volatility and also present huge questions around the future liquidity of these stocks (it can vanish rapidly and unexpectedly). These are some of the major factors suggested in 2011 ‘flash crash.’

Beyond cash and bonds, really great short systems with really tight position controls and risk budget allocations and controls may be the single ‘best’ thing in theory (by that I mean the single system most likely to retain a strong negative correlation with long only stock systems in nearly all market conditions - barring total market collapse), that can be expected to hold up in severe down markets. But there are many, many, many risks in shorting. Most people do very poorly at it, losses can be unlimited and even someone like Buffet stays away. And P123 lacks the tools to do it well at the individual stock level (for example, not true ‘estimates’ of short side liquidity or bost to borrow). So, most of us are better off steering clear on P123.

Having said that, I’ve watched several hedge funds use it ‘out of sample’ for more than 10 years and do very well in using it to smooth their risk profiles.

So, what else can be done:

  1. Allowing the book functionality to a) have leverage and b) force ‘long short’ systems into gross market and sector and industry and beta neutrality (based on user defined inputs), is one thing to consider. But, it requires much better data in terms of truer estimates of ‘cost to borrow’ and likely short side liquidity.

  2. Having the ability to create better backtests on developer / user defined Industry characteristics would be helpful. The industry ranking and factors are still very limited and hard to work with. If for example, I want to find all industries with declining margins and falling sales and increasing consolidation, it’s very hard to do so - so I can’t really learn about them in a systematic way. So, this could be improved to allow better backtesting of user defined industry characteristics to see how those contribute to future industry volatility. It’s very hard often for a company to escape industry turmoil.

  3. Better, more flexible market/ETF rankings on custom universes might help - at a min, I’ve had feature requests for some time, allowing us to use ETF rankings on a specified custom universe in buy, sell and hedge rules. I can create rules that work for me with these conditions and the new custom series - but need to be able to use custom series in ETF ranks.

  4. Better shorting data to account for estimating ‘likely cost to borrow’ and ‘likely true short availability.’

  5. Accounting for taxes in trading systems - to better account for the ‘risk’ of not truly understanding total system costs.

The above are in no order.

While fundamentals and logic are useful tools, there are many ‘purely mathematical’ risk control approaches that have been used by top hedge funds for more than a decade that work well enough, at least to my satisfaction. That’s why the people building them make tens of millions of dollars to billions. We can knock them all we want, but they are not (all) con men - especially pre-fees - they know what they are doing.

As far as all the other notes on what could be ‘good’ and what could be ‘bad’ on individual stocks - this feels like ‘chasing your tail’ - the ‘goodness’ or ‘badness’ of any factor depends on how it performs historically and other market participants agreeing and putting their money in those same ideas.

The simplest thing is what we all agree on - buying companies a lot of other people will want to buy in the future, enough so so that the price will be driven much higher than what we buy at.

In terms of how to operationalize that, that’s what every system is trying to do. In terms of running companies, strong margins are much preferred to weak ones. I’ve run or been on many boards, and this is near universal. Margin slippage is almost always a bad sign, but you hope that you can make it up in volume. I agree, it’s less clear in terms of investments. Really strong companies have great people, great margins, hi growth and early and continuing industry domination in rapidly emerging industries - and sometimes legal or tech based monopolies. These companies are great in theory. They are still often bad in practice as public stock investments, because people tend to overvalue them and their multiples are too high. Only backtesting and history tells us this. Really crappy ‘cigar butts’ may be good - because some backtests and investors say they might be good, but Buffett and many others don’t like them at all and insist on a quality component.

52

Marc - The problem I have with your argument(s) is that the cat is already out of the bag. Everyone has access to R2G in-sample data. Therefore one can’t say “let’s not use historical data in our analysis”. This is like a courtroom lawyer objecting to the relevance to a line of questioning when he himself previously opened the door to it with related evidence. (Yes I watch too much TV).

Now I understand that R2G is going to change and I hope that in-sample data disappears completely. However, in reality this is impractical as in-sample data will still be a necessary part of the book technology. I can’t imagine people creating books with no back-data. I can imagine the uproar that will occur if all that back-data is taken away :slight_smile:

Yes you are right that in times of crisis various assets become very correlated. But in times of crisis, risk is high, so that should not be surprising. When holdings move in tandem, lots of money can be lost, lots of money can be made, but one thing is for certain - risk is high. It doesn’t take rocket science to prove that.

There is much that can be learned from studying the in-sample behavior of models/books that may be applicable to how risky the model/book is going forward. And I believe it is as relevant as the multitude of fundamental factors that are being brought into the picture, all of which need the same level of proof as correlation.

But as I said before, the real value of some sort of correlation analysis is at the book level. Not only for historical data but ongoing so the investor is alerted to potential changes to the correlation of the underlying assets. I know that I am not comfortable putting together books while operating blind. I believe that some sensitivity analysis is required, how the underlying holdings behave under various types of economic stress - interest rates, dollar movement, etc. Historical data is there for that type of analysis. Is it perfect? No. But I think it is better than a pure stab in the dark.

Another thing that I would like to point out. I have communicated with a number of investors using P123, and a lot of them have $500K+, ideally suited for multiple diversified portfolios. These are also the same individuals who can afford the higher monthly fees… I believe that book technology is where P123 needs to put its future efforts in order to truly thrive. Providing tools in order to analyze how books of portfolios react to a multitude of different economic climates is imperative for P123’s future. If I were a wealthy investor, I would demand that.

Steve