Can improve simulation performance by eliminating big loss experiences?

Mgerstein/Paul/Steve/experts,

The sysem don’t know it is a big winner 152% or big loser -62%.

Please, help to share your real world experience on this
Can improve simulation performance by eliminating big loss experiences.

Just would like to get first hand experience; how the out of sample performance if big loss trade is eliminated by analyzing and include a rule to skip;
not by exclude list;

Had experience of holding a stock with 50% loss,
How to treat/accept the loss psyschologically to still stay disciplined with the system.

Believe, your answers will help to clear
one of the barrier/roadblock to committed to the system.

Thank you for sharing knowledge and experience.
Kumar


Can go after big loss trade to skip for improve performance.png

Can you improve simulation performance by eliminating big loss experiences. Yes.

Is that something one should do? Absolutely not! That would be a prescription for out-of-sample disaster.

Big losses are part of investing. That is an unalterable fact. Another unalterable fact is that we can’t predict everything no matter how hard we try. We can data mine to manipulate and inflate the results of backward looking studies, but when it comes to the future, that sort of thing cannot be expected to succeed.

We cope with big losses through the way we model (the more we aim for stability on the downside, the more we also get stability on the upside; we all have to make our own reward-risk tolerance choices) and with our portfolio holdings choices (the smaller the number of holdings, the more exposure we have to big portfolio gains and big portfolio losses).

I always stick to my simple 2 rules when looking to commit real money to a well researched model:

  1. Expect half the Alpha of your model
  2. Expect double the drawdown

If you are fine with these 2 expectations, go for your model.
If not, keep researching.

Werner

Kumar, why not use stop losses?

Firstly, don’t panic. If your system has only a few 60% losers over a 15-year simulation period, you’re doing pretty well. I went and looked at one of my models, Consumer Indulgences, and it also had a 60% loser. In fact, it had two … maybe.

The first loss was SBUX during the 2008 market correction. Bought in 2007, it was sold just as the market was turning around. That’s what happens in models that stay fully invested and I don’t think there’s a way around that doesn’t use market timing of some sort.

The second 60% “loser” was MO. By inspection, the equity curve shows a gain (the transaction in question is the first buy/sell pair). A quick spreadsheet run shows a 128% gain vs the reported 60.6% loss! So what happened? It’s the perennial issue of P123 not including dividends/distributions in the return stats. In essence, the returns reported are price only returns. So check your worst performers and see if they’re really winners after dividends!

For CI, the remaining losers were 45% and less and I’m OK with that since the portfolio max DD is only 35.5%.

So;

  1. make sure your losses are real
  2. if real and you don’t trust market timing, go for acceptance
  3. if you can’t accept the losses, use market timing
  4. if you can’t accept market timing and losses, invest in something else :wink:

Best,
Walter



Kumar,

For what it’s worth, here are my thoughts…

I think it helps to break the system into 3 parts: a) the alpha-generating part which ranks, sorts, and/or filters the universe for the best investing ideas; b) the risk management part which consists of managing leverage, stop losses, sector exposure, etcetera to align with the investor’s risk tolerance; and c) implementation nuances that, while including things that will affect real-life performance, the investor is indifferent about due to either ignorance, ambivalence, and/or perceived inconsequentiality.

The better your system is at alpha-generation, the less you need to worry about under-performance over the long-run. Having confidence that your system generates alpha is, I think, the best way to gain that psychological fortitude to withstand draw-downs and large losses. Having that confidence also helps with risk management – if your underlying system is good, less thought and effort must be spent tweaking the risk management parameters in order for the system to have achieved historically satisfactory results. Note: if you are tweaking your risk management parameters to achieve good back-test results, I guarantee that you are over-fitting the data and that past performance will not be indicative of future. Moreover, the better the underlying system is, the more inconsequential the implementation nuances will become.

Hope this helps.

Mgerstein/Werner/Walter/Primus,

Thank you very much for sharing experience to follow quant system with discipline and accept the big losses as part of investment business.
It helps me to think and work within boundary.

Thank you for generous helps to sharing decades of knowledge; it saves me lot of time.

Thanks
Kumar

PortfolioPerfection,

Believe, Momentum system requires stop loss (SellRule: GainPct<-10).

For Value system the rank will take care of stop loss.

GainPct< -10 pct stop loss performance less than rank only stop, still it leads to -50 pct down in a stocks (Please, refer the screen shot).

As Peter Lynch, Warran Buffet systems are value system; their system will stay invested in draw downs without stop loss.
IBD50 is growth and momentum system; it may need tight stop loss.

Here, screen for illustrations.

(Note: i have make it as full my smart alpha model; will wait for 3 more months out of sample).

Thanks
Kumar



Mgerstein/Experts,

Expert’s comments using STOP LOSS (at Sell Rule: GainPct < -10) on deep value system is appreciated.

The stop loss is executed on Monday in the attached simulation screen using the Sell rule.

Thanks
Kumar

Hi Kumar,

I don’t think Stop Loss is a valid way of reducing your big losses, and here’s why:

Imagine there is an earnings report at 5 pm with bad news. You have a -10% stop loss and the stock opens the next morning with -20% lower price compared to the previous day. Your broker will execute the trade following the stop loss rule, but the only price the system can get is the -20%. So the stop loss didn’t work…

Even worse, imagine the stock rises back to a level -5% at the end of the trading day. Then you have sold the stock at the worst time, while a hold strategy would not have created nearly as much damage.

Testing I’ve done in the simulation confirms that stop losses, although great discussion topics in seminars and blogs, don’t work in real-life markets. Sevensisters’ post above outlined a very realistic scenario, one that happens a lot.

You can never ever never never never protect yourself against big losses in a stock through a long-only model. You can get some protection on a portfolio level through a model that aims at low volatility factors and has reasonable diversification.

You can get protection on a portfolio level through broader asset diversification (although you can’t use testing to help you evaluate this because the great diversifier of the past 35 years, fixed income, no longer has the same role – but it can do so in different ways going forward – flat or moderate declines as opposed to steady gains).

You can probably do more to protect your downside with market-neutral long-short modeling, where you aim to remove the market and profit from factors; that can be evaluated in the simulator. That can be important because a significant amount of the big volatility comes from the market, as opposed to stock-specific issues. This won;t protect you against the occasional company-specific thing, but with proper diversification, a decent number of holdings, and a good market neutral model, you can accomplish quite a lot at the portfolio level.

And of course you can enhance what you do on p123 through off-platform activities such as use of options; i.e. buy puts on potentially volatile stocks you own. You’ll cut into your return (through the premiums you’ll pay to get the puts), but you can recoup part or a lot of a 50% loss with profits on rising puts (but this will require a lot of off-platfdorm analysis lest you spend so much in puts that expire as to nullify your positive stock returns).

And you can experiment with on-p123 hedging; as with long-short, it works best against market volatility rather than stock-specific, but if you can’t/don’t want to deal with the extra work of options and don;t want to hold a ot of individual short positions this is something that can help you. (I’m talking about permanent always-on hedges, not market timing unless you have a model yopu beleive can be truly forward looking and not simply predicting the 2008 past).

Thank you for your detailed comments on stop loss usage.
My system not using stop loss as average holding time is between 3 to 4 months. :slight_smile:
More holding time to recover.

I don’t want to explore option as it is high-risk game. many complex strategies; not just buy and sell.

Few months ago, i have started to experiment dynamic hedge 10% TZA. Here, screen shot for reference.
I will continue to work further.

Thank you for open ideas to stay with big loss position and reduce draw down in portfolio using a hedge.

Thanks
Kumar


This is exactly how I look at hopeful results from a model. I never commit a dollar unless I’m content with accepting half the Alpha and Double the DD.

As for filtering large losers. That’s not an easy answer. The way I see it is that you have to address where and when your losses occurred. Stats alone don’t tell you the entire story. What events may have triggered the loss, was it foreseeable? Did any of the other losers suffer from a similar situation? Was it market timing that held the brunt of the responsibility? There are many reasons why you end up holding a large loser but I also agree with Marc that sometimes that’s just part of the game and accepting large losses when everything else looks copacetic is something we all face. Hence the saying “don’t put all your eggs in one basket” :slight_smile:

I am currently following 64 stocks (& a few ETFs) in 11 Ports used in my 5 accounts (IRAs, Joint, etc.). I ALLWAYS use stop losses. However, not to maximize performance, but to minimize drawdown and to protect capital.

In 1973 I was invested in a few mutual funds and 20 or so stocks. Due to the oil embargo and following recession in 1973 & 1974 the S&P 500 lost 48%, and I lost 35% in my funds but over 60% holding small cap stocks before I got out. It took me 2 1/2 years to recover (I missed the first 20% market rise before I got back in). Since that time I have used stop losses to protect capital.

That saved my butt in 2001 on 09/11 when many of my MicroCaps fell between 60 and 80% in less than an hour. I had a 30% stop loss on all my holdings and they were stopped out between 35 & 45% before I could even get my computer turned on (prices were falling so fast that the prices I got were below my stop). I was then able to sell the rest of my holding before they closed the market for trading that day. A loss of 40% requires a gain of 60% to recover, but a loss of 70% requires a gain of 233% to recover. The first will take a few years to recover, but the second will take over a decade.

So if using a stop loss causes me to forfeit a few % on my annual return, I’ll take that while trying to save my butt during the next major world disaster.

Denny/Mgerstein,

Believe, SP500 universe works different from Microcap.
Value stocks works different from Momentum stocks.
P123 stocks selection works different from other stocks selection system.

I am using SP500 price series and SP500 EPS trend to time the market; (no economic indicator, no individual pct gain stop loss).

It looks like my system works around 20% to 25% draw down; it recover 100% within 3 to 6 months each time after the draw down.

I will start another thread to compare our best model performance around 09/11/2001 event. :slight_smile:

Thanks
Kumar

Instead of stop-loss orders on individual holdings (which seems geared towards idiosyncratic risk protection), how about addressing systemic risk with deep out-of-the-money put options on something like the SPY?

Walter

Much to my surprise, I’ve never found that any kind of stop loss increased performance in any way for any of my models.

Walter,

We can’t test option strategy; it requires some experience.
Hedging the model with 10% inverse ETF FAZ (strongest candidate in bear market) dynamically will be easy to follow and will have a p123 backtested edge.

I don’t have any experience in option; the only knowledge i have is; it is not for beginner;
and high-risk activity and stay away;

can illustrate with example how to peroform
deep out-of-the-money put options on something like the SPY ?

scenarios:
I buy and hold 100 SPY for 100 dollar = 10,000 dollars.

How to take option to protect this 10,000 investment (before taking put option; i am assuming all 3 unexpected scenario).

Scenario1:
SPY stay the same = 10,000 (Current investment remain) in 3 months because of unexpected flat market.
Scenario2:
SPY down 10% = 9,000 (Current investment become) unexpected downtrend in 3 months;
Scenario3:
SPY up 10% = 11,000 (Current investment become) unexpected uptrend in 3 months;

How the Just same put option will work in all 3 above 100% unexpected cases.

Thank you for sharing knowledge and experience.
Kumar

The inability to simulate option strategies in P123 shouldn’t preclude their use.

https://www.thefelderreport.com/2016/08/15/worried-about-a-stock-market-crash-heres-how-you-can-tail-hedge-your-portfolio/ covers some of the mechanics of using puts for tail-risk hedging. Curiously, it doesn’t mention portfolio beta. There are other articles on the CBOE website that do, however.

Since I’m working my way towards tail-risk hedging my portfolio, I’m curious about the general opinion of this strategy. The obvious downsides are initial setup cost and the cost and time of rolling the position each month. But what else?

Best,
Walter

Walter,
You are obviously an advanced trader. There are good reasons for protection for advanced traders.

But the good reasons are pretty complex involving volatility drag, optimal Kelly or plans to take money out of your account in the near future. Usually there is leverage involved. The leverage increases the volatility drag or puts you above optimal Kelly.

For many–including me in my SEP-IRA account that does not allow leverage—making a bet with negative expectations just reduces your returns.

It is like taking insurance in blackjack. It might make sense to take insurance if you got overly excited and have bet most of your stack. Otherwise, if you are counting cards and making smaller bets (less than optimal Kelly), you should only take insurance when the count says the odds are with you on the insurance bet. The insurance bet should be looked as a separate bet unrelated to your other bets.

That is the specific advice that Ed Thorp gives in his book “Beat the Dealer.” But then again, Ed Thorp made most of his money from the mis-pricing or warrants. He did short stocks while buying warrants for the same stock. So there is a place for this.

It may be a good idea for you. But most retail traders do not have enough of an edge to throw money away on bets that are expected to lose.

-Jim