Back to Community
VALUE Factor Composite - Feedback requested please

Hi All,

Here's my 'pure' VALUE composite strategy, with my own way of defining 'Value' as a factor (hint: it's not low price to book ;)). Please let me know what you think!

Loading notebook preview...
16 responses

And here's the alpha decay and risk exposure analysis.

Click to load notebook preview

Thanks Antony!

I hadn’t actually. It’s ‘naturally’ slightly negative book to market value. Do you think that’s a concern?

Solid! Great job as always, Joakim.

Nice, work Joakim! The alpha decay (although not much "decay") is very consistent. I like how you put your own spin on a value factor. We can see that in your risk exposure analysis and common style returns. I knew you'd come back with something new / different!

Thanks guys! My biggest worry (as always) is that it’s overfit.

Working on a GROWTH composite now and would like to do a QUALITY one as well eventually.

Definitely very stable! Look forward to seeing how it holds up OOS.

Thanks @Viridian, me too! (fingers crossed)

Here's the tear sheet for how it performed during the hold out period. Unfortunately it appears to correlate heavily with dog sh*t OOS, just barely meeting the requirement of being profitable for the last 2 years rolling... To be honest, I'd be surprised if it's overfit (the factors are pretty general I think), but it certainly appears that way. I plan to leave it running in the contest for at least 63 trading days though (assuming it doesn't fail the profitability requirement), hoping it'll at least be uncorrelated with my other strategies.

Click to load notebook preview

Our strategies will always be overfit to some degree. That's unavoidable. The question is whether the remaining signal is adequately strong. Your OOS drawdown is twice your in-sample drawdowns, but -4% isn't terrible. It starts struggling during the 2019 recovery that followed the Dec 2018 correction. With valuation as stretched as they are, I would expect VALUE to do some really funky and volatile things at this point, especially when there's a lot of fear in the market of crash and when that fear subsides and the market decides it's still in growth mode and rates get cut anyway. Wild times. The projection curve isn't taking enough history into account, so I would take that with a grain of salt. I would definitely give it more OOS time.

Well, there goes my VALUE Factor Composite. It was withdrawn from the contest on 23rd of August for not meeting the 2 year 'profitability' requirement.

I find it a bit unfortunate because I don't actually believe this one is very overfit (or p-hacked/data-snooped, etc. for that matter) and I think it might be mostly due to 'Value' in general performing poorly this year. I suffer from massive Joakim-bias though, so I could definitely be wrong on this one.

For now, I will try to find a few 'Value' factors in the Estimates dataset (currently not included in this composite), to see if that could help the strategy to remain profitable during the recent period. I'm also hoping 'Value' in general will recover a bit soon, so it will again qualify for the contest and I can then resubmit it. Fingers crossed.

Still performing better than a standard value portfolio.

Clone Algorithm
Total Returns
Max Drawdown
Benchmark Returns
Returns 1 Month 3 Month 6 Month 12 Month
Alpha 1 Month 3 Month 6 Month 12 Month
Beta 1 Month 3 Month 6 Month 12 Month
Sharpe 1 Month 3 Month 6 Month 12 Month
Sortino 1 Month 3 Month 6 Month 12 Month
Volatility 1 Month 3 Month 6 Month 12 Month
Max Drawdown 1 Month 3 Month 6 Month 12 Month
import quantopian.algorithm as algo
def initialize(context):
    algo.schedule_function( rebalance,algo.date_rules.every_day(),algo.time_rules.market_close(minutes=30))
def rebalance(context, data):
    order_target_percent( symbol('SPYV'), 1.0)
    order_target_percent( symbol('SPY'), -1.0)
There was a runtime error.

Thanks @Viridian! For some reason your backtest doesn't load fully for me (computer says 'No!'), but I could at least see the 'Overview' returns graph.

Here's my second stab at a VALUE Composite, which I'm (re-)submitting to the Q Contest. I hope this one doesn't fail the profitability requirement right out of the gate as well.

Click to load notebook preview

And here's the Alpha decay / Risk exposure, etc. NB.

Click to load notebook preview

That one did even worse during the holdout period, and failed right away. I'll leave the Value one be for now as I don't want to start 'training' too much on the holdout period. I might look at trying to resubmit one of them (probably the first one) at a later stage, when hopefully 'Value' has started to recover. In the meantime, I'm working on different ways to combine all of three of the different factor composites (Value, Growth, Quality) into a single strategy. It's looking promising, but maybe this would be against the new guidelines? Anyway, I'll start a new thread for that one once I feel it's ready. Hoping to submit it to the contest in place of my 'Value' slot.


It looks to me your algorithm was on live "regime change"!

Interesting reading Value vs Momentum ±5σ returns and market news in recent weeks.. an exodus from "growth" and "momentum" stocks to "value", dividends and gold in times of global uncertainty.

Hi Karl,

Thanks for your comments. Yes, quite interesting times in factor land recently. I do think Value is relatively cheap, both compared to historical 'Value' spreads, as well as compared to other factor types/styles. I therefore wanted to have my Value composite running in the contest as well, and with the recent bounce back for Value (might mean-revert again soon though), I was hoping my first version would now meet the 2 year profitability requirement. Turns out it does, so assuming it doesn't fail this criterion again, I'll leave it running in the contest for at least 63 trading days. I now only have 1 slot left in the contest, and I haven't decided what to use that slot for yet.

PS: The only change I made to my initial version was to make it more concentrated (150 long, 150 short), bringing up average volatility to around 2.5%, and less likely to dip below the 2% floor in the score calculation.