Tricks to reduce drawdown?

I am designing a strategy and I want to reduce the drawdown without adding stoplosses. Is there any trick? Like for example increasing the number of positions, or adding a particular factor in the ranking system.

There have been many discussions over the years in this forum about this.
Do a quick search.
Here is an example.

In general sentiment factors seem to have the most impact.

I used to have part of my portfolio in defensive stocks strongly inspired by the Utility/Consumer Staples strategy from ‘What Works on Wall Street’. You can read about it here: .
Very briefly, the strategy invests in consumer staples stocks with high shareholder yield, and utility stocks with low valuation, it has quite low drawdown but still good returns.
Another option might be low beta factors or similar

1 Like

My portfolio is an IRA in withdraw mode and one of my management tilts is trying to limit drawdowns. I use the Screener where I have multiple small count strategies which I combine together. Priority strategy selection criteria is a relatively low drawdown and (very) high Sortino Ratio.

Depending on what else is in the strategy, these additions may be positive, negative, or neutral. And the results vary from strategy to strategy.

Eliminate an entire Sector (or select only some sectors) can help quite a bit. I have found that removing Energy helps (hurts) quite a bit.

Using FRank with #PREVIOUS against a factor towards the end of a screen to trim off a small portion from one end or another or both. Short Interest related factors have been helpful here, but be careful here because a real low value may mean it is essentially impossible to short and that tail of the distribution needs to be trimmed too.

The current Selling Short / Buying Puts thread has generated some productive factors for eliminating losers which I’m just starting to work through. Trimming the bottom 5% from the SGandAGr%PYQ factor dropped the working set by 2, reduced DD from 28%+ to 21%-, and increased alpha by 6.5%. But on two other screens it had no impact.

I also use some tactical market timing to step out of the market where backtest has indicated a positive impact. The form is: Between(SMA(8,0,#BENCH)/SMA(15,0,#BENCH),0.98,1.0135)
Lower bound range is (0.96 0.99). Depending on intermediate market conditions one needs to change the values. This gets you out early on a local top, but also in early afterwards. The screen I pulled the rule from has a DD just less than 10% over the last three years while being in the market 76% of the time and an alpha of 63% versus R2000 Value. The downside is that it also has a 65% semi-weekly turnover rate because of other technical rules – trade-offs.



Why don’t you still do that?

Good question, I would rather just focus on as high returns as possible and accept drawdowns. A big drawdown is also a great opportunity. After a market crash you can buy stocks that have fallen a lot and this will usually mean a swift recovery

I would not go that route, the lower the drawdown, the higher the probability you overfit to the data. I like systems that have a 50% DD and have > 50% annual return with a at least 0.7% correlation to the Spy.

Better to time the strategies with an overlay (outside of the system, the stuff that is needed to time the market and your strategies is not in the database of P123!).

Here and example of timing stuff…

Best Regards


Thanks a lot for the replies.

I pay attention to drawdowns in backtests because I find that it reduces overfitting.

I find that when designing ranking systems, lower drawdowns in backtests have some correlation to out of sample outperformance.


I agree to what you are saying based on the research conducted in this paper.

Annualized return in backtest show negative correlation to out-of-sample.
while Annualized volaility (R^2 0.67) and Max DrawDown (R^2 0.34) shows substantial IS-OOS predictability.


Here is the screenshot.

Here is the link for the review on the paper :

To clarify: In my testing, I found that maximum dd in sample for ranking systems was correlated to oos returns.

This may or may be not be what the paper says, but it doesn’t change my experience. I didn’t read the paper but likely the study design was different in a way that may change the correlations between ins and oos.

For what it’s worth my current thinking on the topic (learned from quite a few mistakes) is to limit drawdowns by diversifying across different strategies and asset classes. I do not recommend trying to time the market at all (this has cost me significant gains). My current portfolio allocation looks basically like the following

20% - Micro Cap Value + Quality
20% - Large Cap Momentum
25% - Consumer Staples
25% - Healthcare
10% - 2x Levered Long Duration Treasuries

I rebalance on a 12 month rolling basis. Basically divide my port into 12 tranches each of which is rebalanced every 12 months, so I am rebalancing 1/12th of my portfolio every month. My goal is to keep my max drawdown around 40%.

Do you mean the dd is determined under the ranking performance tab? Is that what ‘for ranking systems’ means?

I didn’t read the paper, but I’m assuming that the reason drawdowns show predictability is due to the correlation between drawdowns and volatility. In other words, because volatility is very highly predictable, one would expect drawdowns to be quite predictable as well because volatile strategies produce higher drawdowns.

To Chaim’s point, which I don’t quite understand (I share Walter’s puzzlement over the phrase “maximum dd in sample for ranking systems”), it’s well known that lower volatility stock strategies tend to result in higher returns.