Spread Trading Strategies

[sle]

https://en.m.wikipedia.org/wiki/Stationary_process
That's what I meant by "stationary". Any market neutral strategy is usually based on some transformation of non-stationary asset prices into a stationary series (returns, beta neutral returns etc). Otherwise I am in full agreement with you.

 

[wlnd]

... Spread trading is the basis for most all arbitrage

... What the trader is trying to model and ultimately capture in a proper spread trade is either the convergence or divergence between at least two highly correlated products

... for example, that you thought Chipotle and Yum Brands would significantly outperform other restaurant stocks, and you didn't want your premise to get blown apart by the broader market

...

this is a friendlier explanation

 

[BorsBamse]

Bone,

Previously you mentioned that mean-reverting futures spread strategies haven't worked well in the past years. Why is that?

 

[bone]

Bone,

Previously you mentioned that mean-reverting futures spread strategies haven't worked well in the past years. Why is that?

Fundamentally driven spread differentials will continue to run well past the meandering statistical variance driven type of spread differential. Until the spread differential overwhelms modeled historical trading ranges you cannot tell the difference between the two. These moves are amplified by program trading.

The traditional way to trade mean reversion was to model one or two sigma statistical variances and fade the move. For the past couple decades this has been very painful for not only futures spreads but according to the four Bright Trading clients I have taken on it also became over time a very painful way to trade equity spreads.

Now, let me be perfectly clear - I routinely take a trade entry on a spread differential that has diverged over some extended period of time and I will bet that the spread differential is going to converge (or mean revert). So technically I do trade mean reversion. The important difference is that the way my trading model and entry signal has been designed it will actually require a confirmed change in the spread differential behavior in order to generate the trade entry signal.

In other words, I'm not simply standing in front of a moving train, holding up my arms and yelling "it's a two sigma move - you've gone far enough!". My trading model requires some confirmation that the divergence state has re orientated into a convergence state.

I hope I didn't get too far into the weeds about that. Let me know if that helps.

 

[sle]

In other words, I'm not simply standing in front of a moving train, holding up my arms and yelling "it's a two sigma move - you've gone far enough!". My trading model requires some confirmation that the divergence state has re orientated into a convergence state.

How do you achieve that? Do you look at fundamentals, positioning etc?

 

[bone]

How do you achieve that? Do you look at fundamentals, positioning etc?

I've got over 200 clients who would kill me if I get into tangible, actionable specifics. My sincere apologies.

But you should know that spread differentials can be analyzed, modeled and traded. There's usually more liquidity there than the outright flat price instruments, and any major Chicago FCM is going to be able to clear them, execute them and apply the SPAN margin credit offsets to your account equity and intra day buying power.

If you look at any futures settlement sheet - almost all of the traded volume in expires other than the prompt month is going to be spread trade related.

 

[BorsBamse]

Fundamentally driven spread differentials will continue to run well past the meandering statistical variance driven type of spread differential. Until the spread differential overwhelms modeled historical trading ranges you cannot tell the difference between the two. These moves are amplified by program trading.

The traditional way to trade mean reversion was to model one or two sigma statistical variances and fade the move. For the past couple decades this has been very painful for not only futures spreads but according to the four Bright Trading clients I have taken on it also became over time a very painful way to trade equity spreads.

Now, let me be perfectly clear - I routinely take a trade entry on a spread differential that has diverged over some extended period of time and I will bet that the spread differential is going to converge (or mean revert). So technically I do trade mean reversion. The important difference is that the way my trading model and entry signal has been designed it will actually require a confirmed change in the spread differential behavior in order to generate the trade entry signal.

In other words, I'm not simply standing in front of a moving train, holding up my arms and yelling "it's a two sigma move - you've gone far enough!". My trading model requires some confirmation that the divergence state has re orientated into a convergence state.

I hope I didn't get too far into the weeds about that. Let me know if that helps.

Thank you for your lengthy reply,

So either people have had a look-back period that's been too short (poor modelling) or spikes in spread volatility have become larger/more frequent/less normally distributed across the commodity space? Or a combination of both? To what degree can the yield-seeking behaviour that we've observed in the VIX futures complex be seen in the rest of the futures space?

It seems to me that because of the fundamental, supply/demand driven nature of commodities in general, there is a higher propensity for prices to mean revert. Of course lag times due to seasonality, production cycles, etc, means it does not happen instantly, but statistical measures of outlier events should be reasonable indicators of future mean-reversion? Ie. there are fewer 'regime-shifts' in the commodity space compared to say equities. (I suppose shale was a big 'regime-shift' for the oil markets, but happened slow enough for traders to adapt). Therefore, a disciplined approach of only initiating spread trades at 'true' extremes should be the way to go? Easier said than done ..

I come from a vol trading background and tend to think of all market positions as either being implicitly short vol or long vol. It seems to me that one could to some extent apply similar thinking for commodity calendar spreads? Ie. since there is limited downside to most commodities, and unlimited upside, I supposed one could view a distressed commodity with high divergence as a type of call option (ex. long front month CL, short back-month CL in December last year). I suppose I'm assuming that most of the time back-months are typically less volatile than front-months (although definitely not the case in the NG market lately) but have yet to begin serious modeling.

Most of the true outlier risks seem more difficult to model, such as government price intervention, weather, etc. Are these the the hidden risks that futures spread traders actually get paid for taking (knowingly or unknowingly)?

Apologies if the above comes across as utter gibberish, trying to absorb as much as possible here. Thanks again =)

 

[AbOoJoOudi]

not good for most of traders

 

[bone]

not good for most of traders

I've got a couple hundred clients who might disagree. Generally speaking, spread differentials model and trend smoother than flat price outright names. They are also much cheaper to margin and to hold overnight as compared to outright futures.

I've had many, many independent clients start out with very modest capitalization - you can margin a Eurodollar Futures Butterfly for a couple hundred dollars.

 

[bone]

Thank you for your lengthy reply,

So either people have had a look-back period that's been too short (poor modelling) or spikes in spread volatility have become larger/more frequent/less normally distributed across the commodity space? Or a combination of both? To what degree can the yield-seeking behaviour that we've observed in the VIX futures complex be seen in the rest of the futures space?

It seems to me that because of the fundamental, supply/demand driven nature of commodities in general, there is a higher propensity for prices to mean revert. Of course lag times due to seasonality, production cycles, etc, means it does not happen instantly, but statistical measures of outlier events should be reasonable indicators of future mean-reversion? Ie. there are fewer 'regime-shifts' in the commodity space compared to say equities. (I suppose shale was a big 'regime-shift' for the oil markets, but happened slow enough for traders to adapt). Therefore, a disciplined approach of only initiating spread trades at 'true' extremes should be the way to go? Easier said than done ..

I come from a vol trading background and tend to think of all market positions as either being implicitly short vol or long vol. It seems to me that one could to some extent apply similar thinking for commodity calendar spreads? Ie. since there is limited downside to most commodities, and unlimited upside, I supposed one could view a distressed commodity with high divergence as a type of call option (ex. long front month CL, short back-month CL in December last year). I suppose I'm assuming that most of the time back-months are typically less volatile than front-months (although definitely not the case in the NG market lately) but have yet to begin serious modeling.

Most of the true outlier risks seem more difficult to model, such as government price intervention, weather, etc. Are these the the hidden risks that futures spread traders actually get paid for taking (knowingly or unknowingly)?

Apologies if the above comes across as utter gibberish, trying to absorb as much as possible here. Thanks again =)

There's a very common tendency for some traders to fade everything. A wild strain of contrarian virus runs in the blood. I'm not saying that describes yourself per se, but it's an observation on my part that there are predilections that can get a trader in trouble - or at the very least, shut the door on opportunity.

Spread differentials between very highly correlated products (usually intra market but some inter market to a lesser extent) tend to model and behave much differently than outright flat price in the same name. The trading ranges are narrower and the price action markedly less choppy. Frequently, but not always, the delta directionality of the spread will differ from the flat price outright contract (spread construction is very important in this regard). I tend to prefer, and I teach my clients to favor, spread combination constructions that have most of the delta directionality engineered out of the combination. For intra market spreads we tend to favor spreads with a bit more complexity ( butterflies and condors ) and a bit further out in the curve. For many of the uninformed, to them futures spreads are prompt month versus second month calendar spreads - and there is SO much more available than that old tune.

If a trader is just waiting to time a particular spread combination for a mean reversion opportunity, IMO that trader is missing well over 50 percent of the trade entry possibilities with that spread. For example, quite literally, you could take 25 full tics out of a Eurodollar Futures Condor on a divergence bet over a period of three or four months let's say - before that spread even hints at slowing down and reverting back to some previously demonstrated long-held trading range.

IMO the reason that in the past so many traders would look for mean reversion opportunities in spreads was because of the abundance of spread combinations with ample liquidity - thousands in the CME product suite alone, and the simplicity of a one or two sigma fade strategy. But spreads in general tend to trend more than they used to - it's not a simple stat arb exercise any more. Those micro stat-arb dislocations (like OTR 5 year cash notes versus ZF or a basket of stocks versus an ETF Index) get pounded mercilessly by automation to the point that execution slippage outweighs booked profits. I have seen it so many times. And I have seen so many big spread traders from the floor get smoked on the screen. The forward curve is all about supply and it is overwhelmingly Commercial order flow. Just because I model a one or two sigma vol range over two years is meaningless to Louis Dreyfus or Cargill or PIMCO. If a trader takes a contrarian view in a strongly trending market dominated by Commercial players based upon modeled historical norms alone - that's the path of pain. I could not and would not allow my clients to trade that way. Enduring weeks of draw downs for the sake of worship before the altar of one or two sigma isn't worth it. And what's more I'm of the opinion of that to be outdated thinking that is unnecessary quite frankly.

I hope that gives you something to think about.