When we ran Carver’s Strategy 9 with 15 Instruments, we noticed how most of the returns were driven by crypto. However, that had only the three big coins – BTC, ETF and SOL. Since hand selecting instruments to trend-follow is also a form of overfitting, we expanded the universe to include all x-USDT coins listed in Binance since before the year 2019. There are 21 of those.
Once you expand the universe, the sheen wears off.
While the highest returns came from using a Binary Long-Only Equal-weight strategy, it came with a 60% drawdown, ruling out leverage.
Digging into the coin-level metrics, we see how a fair number of coins have negative contributions.
While the Big 3 coins had favorable trend-following returns, expanding the universe did not yield a better portfolio.
Our previous post on Rob Carver’s Strategy 9 experimented with four major Indian indices. We saw that only two of them contributed to out-performance while the others dragged.
Can we just run those that worked and throw away the rest?
The whole point of using multiple moving averages is to avoid overfitting. Hand selecting instruments to trend-follow is also a form of overfitting. Carver repeatedly says that his approach works best on a large set of instruments (start with 100 and whittle down.) However, as an Indian retail trader, we do not have many options. Realistically, we can lay our hands on at most 15 different instruments.
With these 15, we played around with: scaled vs. binary x long-only vs. long-short x equal-weighted vs. inverse volatility weighted.
The results are sobering.
Long-only Equal-weight
Long-short Equal-weight
Long-only Inverse-volatility-weighting
Long-short Inverse-volatility-weighting
Of these, only the scaled long-only equal-weight setup looks promising. However, if you look at how individual instruments performed, it is hard to remain unbiased.
In Rob Carver’s Advanced Futures Trading Strategies (Amazon,) there’s a chapter, “Strategy Nine: Multiple Trend Following Rules,” that uses composite trend-following rules to drive a long-short strategy. We explore the strategy through an Indian market participant’s lens.
There are a number of decision points to navigate. Primarily, long-only vs. long-short, binary vs. scaled and equal-weight vs. cost-screen. Some of these are not practical for retail futures traders. For example, you could use the different SMA rules to scale in and out of positions instead of taking a binary approach. However, that implies that each position will need at least a dozen contracts – a capital requirement that is out of reach of a typical retail trader. Another problem is that the Indian market is notoriously expensive to trade. The cost-screen used by Carver throws out a lot of short-term SMAs, making the strategy unresponsive to short-term market moves.
Summarizing NIFTY 50, NIFTY BANK, NIFTY MIDCAP and SMALLCAP indices through Strategy 9:
NIFTY 50 and NIFTY BANK are poor candidates for this system – consistently under-performing buy & hold. The SMALLCAP index doesn’t have futures listed on it – forcing a realistic implementation to be long-only. The MIDCAP index does have futures on it, making long-short possible. The cost-screen version can be safely ignored.
If we decide to use futures for MIDCAP, we need to make sure we don’t blow up because of leverage. The problem with the long-short strategy is the periodic 20% drawdowns. Even at 2x leverage, that’s a capital impairment of over 40%.
However, looking at the equity curve, it may be worth the extra antacid budget?
The next question is, scaled vs. binary?
The scaled long-short version (red line in the chart above) is objectively worse than binary long-short (green line).
So, the version that worked for MIDCAPs is a binary long-short without cost-screen.
For SMALLCAP, given that it doesn’t have listed futures, we’ll have to settle for the binary long-only version without cost-screen.
We interrogated Claude as to the robustness of Carver’s approach. You can read the back-and-forth here.
Our introduction to trend-following posts on Zerodha Varsity (Part I, II, III, IV and V) used tradfi instruments to build a basic model. What if we applied the same principles on crypto assets?
To keep things simple, we’ll pick only two assets: BTC and ETH. The portfolio is evenly split between the two. Since crypto markets are 24/7/365, we’ll divide each day into 24 hour slots and construct a daily series based on the closing prices at each hour. The portfolio is further split into 24 parts each. Each position is an average of a binary trend signal.
Individually, trend-following boosts the Sharpe ratio of each asset.
You may not have captured the absolute highs but you would have avoided the steep drawdowns.
They are stronger together than individually.
Needless to say, leverage in this scenario would be fatal.
It is easier to work with core strategies that fall into one of these extremes:
low volatility / low max-drawdown that can be levered to achieve higher returns without blowing up, or
high volatility / high max-drawdown and high returns that can be managed through asset-allocation and rebalancing.
The worst ones are those that have neither low volatility nor high returns.
Trend-following techniques deliver on (1). However, not everything trends.
Case in point are the Trendpilot ETFs PTLC over the S&P 500 (SPY) and PTNQ over the Nasdaq 100 (QQQ).
The methodology makes intuitive sense.
However, performance is a different matter.
The trend ETFs have worse Sharpe ratios than their benchmarks.
The silver lining is that the ETF versions are better than naïve trend strategies using only SMAs and would work out to be more cost and tax efficient than rolling your own.
However, the naïve versions all have Sharpe ratios less than 1.0 and high drawdowns. This tells us that the indices themselves may not amenable to trend-following and that they belong to (2) above?
Strategy design should follow the “first make it work then make it better” philosophy. If the simplest approach doesn’t work, then adding bells-and-whistles to it is unlikely to make it any better. If something is not trending, then what exactly are you following?
