Forums  > Trading  > Index trading strategy, part II  
     
Page 2 of 2Goto to page: 1, [2] Prev
Display using:  

EspressoLover


Total Posts: 474
Joined: Jan 2015
 
Posted: 2021-03-25 05:02
> That means somebody could use your signal to trade at 3:50 and exit at 4:00 to make 15% p.a. with only 10-minute volatility.

I believe what's happening is that @energetic is still using the 4 PM signal, but executing SPY at the 3:50 price. That would explain why there's such a sharp drop-off in returns. If that's the case, then the 4-3:50 delta probably represents lookahead bias, and can't actually be captured. If this is mostly represents mean reversion in SPY (i.e. buying the dips that are driven by risk-off sell offs), then that might explain why there's such a sharp drop-off when trading a T+0 signal at T-1.

> Who in their right mind doesn't want to make 20%/year? Do many people really have better options? I really really doubt it.

I understand your frustration. I'm actually a lot less skeptical than most other posters here. (Largely because I looked at something analogous years ago, and found pretty similar results.) If you can actually deliver 20%/year, investors should be beating down your door. Certainly that'd crush most hedge funds out there.

But you have a chicken-and-egg problem. Without a track record nobody will give you capital. But without capital, you can't build a track record. Unfortunately, when it comes to credibility, backtests are pretty much worthless. 20%/year backtests are a dime a dozen. The vast majority of capital allocators are not anywhere near as sophisticated as this forum. All the technical statistical talk about careful hyper-parameters tuning and cross validation will just make their eyes glaze over.

I'll second the point made by most of the other posters. Try to boost the Sharpe, either by diversifying across orthogonal return streams. Or trading at higher frequencies. For example, I think you could extend this to single-name equities by reconstructing the VIX equivalent index on specific stocks. Even if the alpha is weaker per name, that would give you 100+ orthogonal streams, which should easily double the Sharpe.

When it comes to bootstrapping a new strategy, there's a gigantic difference between a 2 Sharpe strategy and a 6 Sharpe strategy. For one, the higher the Sharpe the easier and faster it is to validate statistically. Even at 2 Sharpe, it will take years before any investor is sure you're actually generating alpha. In contrast, a 6 Sharpe strategy can be turned on with very tight drawdown limits, and you'll pretty much know in a month or two whether it's working or not.

But more importantly, a high Sharpe strategy compounds at a way faster rate. That means you can start off with a much smaller capital base and quickly grow into a decent size. At one tenth Kelly sizing, a 6 Sharpe strategy will double invested capital every 3 months. Start with a $100k and in three years, you'll have $400 million. (Or more realistically will hit capacity constraints. But the point is, you'll no longer have a shortage of capital or track record.) The higher the Sharpe, the easier it becomes to solve the chicken-and-egg.

Good questions outrank easy answers. -Paul Samuelson

Maggette


Total Posts: 1301
Joined: Jun 2007
 
Posted: 2021-03-25 08:34
"I believe what's happening is that @energetic is still using the 4 PM signal, but executing SPY at the 3:50 price. That would explain why there's such a sharp drop-off in returns. If that's the case, then the 4-3:50 delta probably represents lookahead bias, and can't actually be captured. If this is mostly represents mean reversion in SPY (i.e. buying the dips that are driven by risk-off sell offs), then that might explain why there's such a sharp drop-off when trading a T+0 signal at T-1."

I second ES on that. I really really believe such things need to be explainable before going live.

Energetic paper traded that strategy for a while now and, if I understand him correctly, it worked out the way his back tests suggested (which is good).

@Energetic Is there an option you trade that thing outside of your day job? I mean the strategy could be automated? Set up an IB account and use their API?

Ich kam hierher und sah dich und deine Leute lächeln, und sagte mir: Maggette, scheiss auf den small talk, lass lieber deine Fäuste sprechen...

ronin


Total Posts: 668
Joined: May 2006
 
Posted: 2021-03-25 10:08
> @ronin I think world politics would be a lot more efficient with a few ronin in the mix ;)

Ha - it wouldn't be as entertaining. I think our @energetic here is a rare find.

And, to be clear, I mean that in the best possible way. Rare is good. Not rare wouldn't be good.


> Who in their right mind doesn't want to make 20%/year?

Oh give me a break.

I do actually do this sort of thing for a living. And I do it reasonably well. And I do tend to know what I am talking about.

You are selling spy when vix is high, buying spy when vix is low. You are calibrating for the threshold and the two gearings. Bonus points if you are linking the threshold to some moving average, and you are calibrating how you calculate the moving average and how you link them. Extra bonus points if you are linking the gearings as well, and you got some extra parameters to calibrate there too.

It's not particularly original. It is the summer intern strategy #2 - it's the one summer interns try after they've tried crossing moving averages.

You have done enough fitting to hide the worst drawdowns. You have also undercounted transaction and funding costs. When you put it all back in, you will be left with with spx beta plus noise. When you subtract spx beta, it's just noise.

It's not a strategy. It's a teaching moment.

Which is ok. It's not the end of the world. You are on a learning curve, and that learning curve may eventually lead to a working strategy.

The bigger problem is your attitude. You are like those German drivers who knowingly crash because they are sure they are in the right and the other driver is in the wrong. There is no strategy good enough to survive that sort of thing.

"There is a SIX am?" -- Arthur

Energetic
Forum Captain

Total Posts: 1532
Joined: Jun 2004
 
Posted: 2021-03-25 15:54
@gaj

That "somebody" should be me. Thanks for idea, I'll think about it. But it sounds too weird. Here's the chart though. Sorry for the size again.


No, @EspressoLover, I'm not using the 4 PM signal, but executing SPY at the 3:50 price. I'm using parameters calibrated at 4 pm to trade at 3:55. E.g. I am using VIX(3:55) and futures prices on the same minute as model inputs to generate positions. The very same parameters may give different direction just 5 min later because the inputs are different.

I have a friend who develops completely different type of strategies but he also trades at closing. He reports exact same thing: performance drops off very quickly if he executes earlier. Just a data point.

Yes, of course you're right about chicken-egg problem. I know, I'm just venting out.

Yes, I should diversify. I've been meaning to do it but never found time. I don't quite follow your point about super-fast growth of Sharpe 6 strategy. You mean by leveraging?

@Maggette: No, compliance requires me to declare all my investment accounts and the brokerages report on all my trades automatically.

For every complex problem there is an answer that is clear, simple and wrong. - H. L. Mencken

Energetic
Forum Captain

Total Posts: 1532
Joined: Jun 2004
 
Posted: 2021-03-25 17:36
@ronin

I assure you, no - I swear to you, that your description has exactly zero relation to what I do. But thank you for your evaluation anyway, it is incredibly important to me. Now move on and share your wisdom with someone else.

For every complex problem there is an answer that is clear, simple and wrong. - H. L. Mencken

EspressoLover


Total Posts: 474
Joined: Jan 2015
 
Posted: 2021-03-25 17:57
> I don't quite follow your point about super-fast growth of Sharpe 6 strategy. You mean by leveraging?

For a fixed Kelly fraction or drawdown tolerance, the return to a strategy scales quadratically with the Sharpe. E.g. doubling the Sharpe quadruples the the return. This happens for two reasons. First, higher Sharpe on the same return stream linearly increases unleveraged returns. Compare S&P 500 returns to a timing strategy where 100% predict long or short correct everyday.

Second, higher Sharpe linearly decreases the drawdown on unleveraged returns. An intuitive way to think about this is that as the distribution shifts to the right, a smaller proportion of the variance falls below the zero-bound. For a fixed drawdown tolerance, leverage scales linearly with Sharpe. Increasing Sharpe is like double-dipping- you increase both the returns and the leverage at the same time.

For example with a 6 Sharpe daily strategy, and a 10-year max DD tolerance of 30%, you'd target 377 bps of daily volatility. That earns you 137 bps of average daily returns. At that clip, you're doubling your capital every quarter.

> E.g. I am using VIX(3:55) and futures prices on the same minute as model inputs to generate positions. The very same parameters may give different direction just 5 min later because the inputs are different.

Hmm... That's pretty interesting. In this case, I'd definitely take @gaj's advice and try shorting the signal between 3:55 and close. At the very least it should be pretty enlightening. The sizable drop-off can only be explained two ways. One, the market moves strongly opposite to the signal in the tail-end of the intraday session. Two, the signal tends to move sharply in the last N minutes, and this move is highly correlated with overnight performance.

In the former case, you can use short the signal intraday to capture a high-Sharpe high-frequency alpha. In the latter case, you should be able to improve the performance by conditioning on whether the signal formed in the last N minutes. I.e. signals that suddenly switch just before close have much stronger overnight alpha.

Good questions outrank easy answers. -Paul Samuelson

Energetic
Forum Captain

Total Posts: 1532
Joined: Jun 2004
 
Posted: 2021-03-25 19:01
> higher Sharpe on the same return stream linearly increases unleveraged returns.

I think of it the other way probably because I'm not optimizing Sharpe but OK, it should be equivalent. Still, this is true under some assumptions that don't hold in my case: killing the long strategy reduces CAGR to 20% but Sharpe based on annual and monthly returns and vols increase. However, daily distribution is more "normal", I guess, because daily Sharpe actually falls a little.

As a footnote, my friend reports that for him optimizing Sharpe in-sample doesn't carry over to OOS. But returns do.

OK, I guess I need to work on shorting. Many thanks!

For every complex problem there is an answer that is clear, simple and wrong. - H. L. Mencken

Energetic
Forum Captain

Total Posts: 1532
Joined: Jun 2004
 
Posted: 2021-03-26 15:41
Short didn't work, I need to dig deeper into that.

For every complex problem there is an answer that is clear, simple and wrong. - H. L. Mencken
Previous Thread :: Next Thread 
Page 2 of 2Goto to page: 1, [2] Prev