I have built a long-term algotrading strategy, how to bring out the most?

[DouglasP]

Would it be feasible to achieve this by using a min-tracking error optimization around the original portfolio (including a short side, be it index futures or actual single stock positions), with beta/factor exposure constraints? That way you'd try to remain as close as possible to the original idea.

[Tralek23]

Thanks for all the replies, I’ll try to sum up on one post.



@EspressoLover

>> Have you tried neutralizing the beta…

Yes, I tried but my strategy operates mainly (but not only) on bullish market phases, so being beta neutral is not beneficial to the strategy.



>> but oftentimes there's a (nearly) free lunch when it comes to reducing lunch.

Sorry, I can’t fully understand the meaning, what do you mean by reducing lunch?



@NeroTulip

>> Have you looked at your strategy's alpha, beta, and correlation to S&P?

Here some stats I’ve calculated:

beta 0.51

alpha 0.061

volatility 0.103

correlation 0.73





@rickyvic

Thanks for your useful suggestions, I’ll pm you in the next days,



@svisstack

Currently I operate only on the stock market, anyway I'll contact you in case I'll create a crypto one.



@gaj

The strategy performs about a trade per week, it is a long term strategy but the positions are not held more than a few days.



@DouglasP

I’ve not tried the mean-tracking optimization. Anyway the strategy signals are not overlapping so I’ve never focused too much on portfolio optimization strategies.

[Its Grisha]

> it is a long term strategy but the positions are not held more than a few days



If you are holding positions for only a few days, transaction costs (both explicit and implicit) become a very serious consideration. Plenty of alphas disappear after accounting for them.

[Its Grisha]

edit: sorry double post

[EspressoLover]

@tralek



(Sorry, previous comment had a typo should have said "reducing vol" not "reducing lunch")



Effectively you don't have a 0.9 Sharpe strat. You have a 0.6 Sharpe strat (alpha/volatility). The additional 0.3 (and lower drawdown vis-a-vis the market) comes from diversifying its exposure with the market portfolio. But that doesn't "count", because nobody is going to pay you to provide beta. If a big investor wants long exposure to the S&P 500, she's going to do it through Vanguard at 5 basis points. Or Bridgewater at 100 basis points. But definitely not you at any cost.



I really want to emphasize this, because the only chance you have of selling this to a serious investor is as an absolute return product. And a 0.6 Sharpe absolute return product may be sellable, but not unless you have a pedigree. It is absolutely essential that you boost the Sharpe to at least 1.0, if not higher.



You mention that the strategy only seems to do well during bullish expansions? Are you sure that's not just because of the long-only nature of your positions? If you think of what you've been trading as a mixture of S&P 500 + [your unique signals], then maybe it's the S&P 500 that's making it bullish-bias. Try backtesting again, but beta-neutralizing every position. (I.e. short an equivalent amount of SPY on every long trade).



Beyond this, I see three likely avenues to boost the Sharpe. The first is turning over your positions more frequently. You say you hold a few days, but why? How frequently are you rebalancing the portfolio? If you're just using end-of-day data, it's possible that intraday rebalancing may reveal more trading opportunities.



Have you evaluated what the alpha realization curve looks like? It's frequently the case that 50% of the alpha realizes in a few hours. If you can exit trades much faster and still get most of the profit, that means less exposure and/or more capital available to do other trades that would otherwise be tied up in long-term positions.



The second option is to simply diversify across a larger basket of trades. I assume that your signals gives you something like a ranking of stocks in your universe? And then you go long the best stock? What about buying the top 5 names? Or top 100? Or all the stocks with positive signals in proportion to the signal magnitude? If you're long-only that doesn't help that much, because you still have market exposure regardless. But if you're beta neutral, then diverisfying the single-name exposure is a big win.



Also, have you evaluated the signal in terms of the short-side? What if you hedge the beta by shorting the bottom ranked stocks (instead of shorting SPY)? Now you're getting an alpha both from the short and long side. In most anomalies the short-side has larger alpha than the long-side. That would naively double the Sharpe. (Not completely because single names have higher t-costs than index hedges, cost of borrow, etc.)



Third, can you expand the size of the universe? Right now you look at S&P 500 single names. Can you do the Russell 3000? Can you add ETFs? Can you expand internationally to Europe, Asia or emerging markets? Commodities? Currencies? Bonds? If you can find 4 different markets or countries with orthogonal performance, you've just doubled the Sharpe ratio.



Finally I want to second @gaj. I'm not convinced that the strategy's performance isn't spurious. I don't want to be negative, but any potential investor will ask the same thing. At 0.6 Sharpe with 12 years of history, the null hypothesis barely clears statistical significance of 2.0 t-stat. And that doesn't account for any lookback bias. Was this the very first thesis you tried? Is any of that performance using in-sample trained parameters? Did you tweak the strategy parameters based on historical performance?



This is a secondary, but very convenient benefit of higher-Sharpe strategies. They can be statistically validated with less historical data. The upshot is if performance is spurious, then all of the previous suggestions will tend to make the strategy look worse. If the strategy is real, then diversifying and hedging will distill the signal. But if you've just overfit noise, then removing any of the vol will just dampen the noise.



This is one of the biggest pitfalls even seasoned practitioners fall into. It's tough to call it when a strategy you've worked hard on ends up being a dead end. But it happens all the time. A tell-tale sign is when performance seems to evaporate when you make any changes. Real signals are usually robust. It's way too easy to be defensive and declare "my particular strategy just doesn't work in a beta-neutral context" or "it has to be rebalanced at market close" or "it just doesn't work outside the US". In reality, you should probably interpret it as a sign that there's just nothing there.

[sharpe_machine]

> Or all the stocks with positive signals in proportion to the signal magnitude?



One interesting question is what to do with rankings when you have the majority of the liquidity outside of top10 (or whatever border you like) positions and alpha is skewed towards top-ranked stocks.

[wquant]

I'm interested in how this would work in practice. Let us say you had a solid strategy, what's the best way to garner interest with a hedge fund (assuming no existing contacts)



Are there firms out there actively looking to seed early-stage strategies or be otherwise helpful in terms of data/ordering infrastructure?

[NeroTulip]

@Tralek23: I don't want you to think I am not responding to you, but EspressoLover has made my point and much more. Nothing to add, except reread his post and get to work.

[svisstack]

BTW. Escalated this to the QuantConnect CEO as we have a contact to check if this misalignment between alpha measurement is on purpose or not. It may be a bug in the QuantConnect that you can't monetize some type of edges (long term mid-Sharpe).

[Tralek23]

@EspressoLover I would like to thank you for your long and comprehensive reply, I hope it could be a helpful resource for others in the same situation.



@Its Grisha

In my algorithm, I account for commissions, bid/ask spread and I trade liquid assets only so that the total slippage will not impact too much on the strategy.



@NeroTulip, ok thanks for your suggestions in any case.



@wquant

>> Are there firms out there actively looking to seed early-stage strategies or be otherwise helpful in terms of data/ordering infrastructure?



I’m contacting small fund and asset manager that are showing interest in the strategy.





Thank you all, you gave me a lot of ideas and feedback, I’ll work on it and I’ll update this thread in case of relevant developments.