- This topic has 11 replies, 5 voices, and was last updated 6 years, 10 months ago by
.
- Posts
I know the topic is generally dismissed, but since the typical hedges haven’t done a thing in the past week or two, I’d like to know if there is a way to program in “cash out” rules for both upside and downside swings that exceed a certain threshold. When everything is going down at once, should we just sit it out until the end of the month, or is there a way to rationally add in stops and back-test their effect?
Ok, so the premise of the whole momentum trading system is that there are asset classes that are anti-correlated over a monthly timeframe. That seems to work, except when it doesn’t, like in the last week or two. Now, what if the QT software had a way to test for the trailing correlations between your hedging assets. So, you designate certain investments as your hedges, and others as your risk assets, and the software computes their lookback correlation. When the correlation has been reliably and sufficiently negative, then allocate as expected. However, when the correlation is insufficient, below whatever threshold, then we substitute go ahead and allocate according to the normal rules, but we also add a cash stop. The rule would be enabled when both the hedge and the asset move downwards within a single monthly investment cycle by more than whatever threshold for pain we set. Is that a bad idea, or a good idea?
Well, this is good in theory. From what I understand hedges we have are “sufficiently negative” on the average, and when not, they tend to mean revert and return to being negatively correlated. That’s the first problem: when correlation is insufficient it’s most of the time too late to switch to another hedge. The other problem is that we don’t have a lot of hedges to switch to. So I suppose this approach would lead to a lot of false signals. It would go like “Oh, gold is not working for second week in a row, let’s switch to bonds” then “Oh, bonds are not working too good, but gold is looking like a good hedge again” and so on. Lot of whipsaw. Another problem is that we introduce another degree of freedom to our system. And each one we have makes backtests less and less reliable.
I still think there should be provision for a ruleset in QT that would allow us to experiment with intra-allocation trailing stops based upon VIX and movements in the hedge/risk designated assets. At least we could test it. I’ve struggled for quite a while to find a way to create a “cash out” rule, but there are really only 2 settings in QT, “Volatility limit” and “Cash Sharpe Limit”.
Alternatively, maybe it would be a good idea to have a weekly reallocating and much more conservative strategy, but that would only be considered if volatility in the main risk strategy is over some limit.
Can you explain a bit what you mean by “intra-allocation trailing stops based upon VIX and movements in the hedge/risk designated assets”? I may test it, if I understand it (outside QT).
From my experience most stop losses do more harm to revenue than help fight volatility.
I just mean if we could model the existence of trailing stops that would trigger at any time, rather than only at the reallocation frequency. If there were “triggers” that we could design around VIX spikes or non-negative correlations between the assets we define as “hedges” (TLT/TMF/GLD/UGLD) and those we define as “risk” (ZIV/QQQ/TQQQ/SPY/SPXL/Nasdaq100), then we could test the effect they might have. They could trigger a move to cash until the next reallocation on whichever assets they are configured. In the trading platform, they would be implemented as one-triggers-other sales using trailing stops or other conditional orders.
I understand that, when there is a panic drop, limit order stops get passed over and do not execute; conversely, market order stops get killed because your sale is for crumbs. Frank and the other guys have all said to only use limit orders, and this sort of thing is why.
Jim Cramer devoted much of his show tonight telling us that the leveraged volatility funds (he named UVXY, SVXY, and TVIX) are being unwound by hedge funds, who are having to sell off stocks because of large inverse volatility margin debts when the VIX took off. (IMO, the guys were wise to specify ZIV.)
That precipitated the whole mess, in his opinion. He was a hedge fund manager and knows how this stuff works.
His advice was to watch the volume on these ETN’s and that, until volume declines and normalizes, avoid adding any positions.
Tom, I understand now, but I still don’t think one can create a robust strategy like that. Your proposal is adding a several parameters to strategies. Eg. what level of correlation would cause a move to cash, how long should it take, do we allow correlation to be high when risky asset is rallying (would you sell everything if both SPY and TLT are going up?) Then you have to test this, and we have very limited testing period, composed mostly of bullish market. For simple “legacy” SPY/TLT strategy we have 2 principal parameters (lookback, volatility attenuator). Your proposal is to add at least another 2 (lookback for correlation, correlation level that triggers stop). Overfitting is hard to avoid. Especially when considering that this “cash everything” events would be rather rare and during our testing period we’ll have just a few (3? 5?) occurrences.
Even now I’m very concerned if some of strategies we trade are sufficiently robust (especially since things are getting complicated more and more, eg. hedge isn’t simple TLT, but another strategy, which is strategy of strategies and so on).
I’m less concerned about market orders because most of our assets are quite liquid.
This is of course just my opinion, I’d love to hear what Frank/Alexander/Vangelis think about it.
By the way, my allocation strategy had 30% CAGR but also a possible 8.85% drawdown–which happened! So the model predicted what could happen. I think that, however I allocate going forward, I will always have in mind how much money that WILL be in a sudden spike down market, and then decide if I am that brave or not.
Another approach would be, instead of using stops and other calculations, a simple “phase transition” that allows one strategy to operate during say VIX < 20 markets, and another to immediately take over any time a simple parameter like that occurs, so we essentially have a simple rule around volatility that forces a “gear shift”. We would only really want to evaluate the gear shift strategy’s outcomes during the times when it is in force.
Of course this can also be done manually, by simply creating two different strategies. I would stick with a monthly reallocation on the primary, but perhaps switch to a weekly reallocation on the high vol strategy. It would still not respond to a single-day event, but it wouldn’t let you sit on accumulating losses for most of a month while you get slaughtered, because neither treasuries nor gold are currently behaving as effective hedges to market risk. They are only dropping “less quickly”. In this case, the only remaining “hedge” is a zero-loss asset, cash. So, we need a better way to introduce cash as an explicit stop-loss in cases like the current market.
I just read an article that compares the current environment to the 1987 and 1994 events, when there is a significant market correction, but bond yields are actually increasing, and therefore bond prices are also dropping. This is a very rare combination that isn’t accounted for in any of the QT models, because it breaks all of the assumptions the hedging approach is based upon. If ZIV were to be closed by Credit Suisse in the same way they’ve now closed XIV, the typical expectation would be for TMF to make up some or much of the loss. That has completely broken down.
If a VIX spike to 30 was sufficient to completely crash XIV, I am not certain ZIV is really a suitable investment vehicle at all at this point. We might be much better off focusing on SPXL/TQQQ or Nasdaq strategies. Unfortunately, the triple-leveraged growth strategies are amplifying market losses well beyond the actual correction, because the hedges in them have completely failed at the same time.
Therefore, I need an alternate strategy that takes over and uses a nicely constructed set of rules that prefer cash and reallocate on a perhaps weekly basis. It would avoid using explicit stops, but it would require some programming effort in QT to reflect as a single parent strategy with an overall performance value.
Some thoughts.
Black swan events will always happen in the markets and they will always be hard to avoid even using intelligent systems. This is because as players adapt to the market, the market adapts back. The next crash is usually different than the last one. Building systems to avoid, or even profit from these crashes is possible but these systems cannot be profitable, long term.We cannot always win. History shows that exiting during these crashes has been a bad idea. Historically it has been better to wait for some type of a reaction and return to short term normally before re-evaluating positions. As you can imagine we cannot advise to ‘stay in’ or ‘cash out’ and It is not our job to do so. What we can do is re-evaluate our systems and see if there can be a mechanism to detect longer term structural shifts (ie, interest rates rising) in order to limit (not avoid entirely) long-term losses.
What Tom proposes has been somewhat used in Wouter’s PAA strategy (implemented here). PAA lost 7-10% so far this February. As Reuptake mentioned, trying shorter term lookback periods will result in multiple whipsaws as modern market protective mechanism tend to correct short term catastrophes (ie mean revert or as Zero Hedge fans call it “The plunge protection team kicking in”).
That said, Frank is experimenting with possible go to cash mechanisms for QT. My own older backtests point to rising rates –> lower market returns.
I agree Tom…While most momentum based strategies pass control of the current trend leader to the next leader, it would be nice to be able to test a “trailing stop” limit going up or a “stop loss” going down. While we can always put in our own $$$ type numbers in this situation, it would be nice to be able to apply an indicator to get out…(it looks like you like the Volatility type indication from your comments), but they could be based on just about anything…. very interesting idea…
Mark
- You must be logged in to reply to this topic.