VP Tactical Cookbook part 2: Trend, Sentiment, Analog
This is the second post of our 3-part series exploring our tactical 1-3 month investment framework where we build an asset’s tactical outlook based on: Trend, Sentiment and Analogs. See Part 1: the key concepts here. The below is an excerpt from our January 2022 thematic report to clients.
While traditional technical analysis indicators are followed by a significant amount of market participants, we wanted to build a set of tools that are intuitive and less correlated with each other.
Technical Analysis indicators are often variations on similar ideas (eg Oscillators). The top table show that 3 very popular indicators (RSI, MACD, Stochastics) on the S&P 500 are very correlated to detrended price and each other.
The VP indicators are less correlated to each other (bottom table). Trend will obviously be highly correlated to price, but the sentiment and analog indicators have a much lower correlation than traditional indicators.
VP Trend: the theory and the backtests
Trend is a common and well known method for forecasting short-term returns with a good track record.
Trend strategies have a convex payoff against index returns. There are two important factors for trend strategies: the magnitude of price moves and number of trend changes.
We have created a Trend Ensemble by combining two trend measures: a custom vol-adjusted momentum indicator and a moving average signal, each calculated over multiple lookback periods (from 1 month to 4 months).
When does this break? It does well when there are large moves and fewer turning points. It does poorly when moves are small and there are frequent turning points.
We split indices into buy and sell groups if trend was positive or negative. We equal-weight these indices daily and this gives us performance for the trend indicator when it is positive or negative.
We also looked at the forward return by indicator percentile, over multiple forward-looking periods. The indicator works well from 1 week up to 3 months.
Trend has good historical performance across all asset classes. However it underperformed from 2010-2018. Markets had smaller moves and more frequent turning points.
VP Sentiment: the theory and the backtests
Most of the time, you actually want to stick with the crowd when the tide is turning from bearish to bullish or vice versa. The only times to be contrarian is when sentiment is very extreme.
When extreme pessimism gets less bad then you want to be a contrarian and buy (top chart). The data is weaker for fading extremely high sentiment.
Our Sentiment indicator tries to capture the psychological effect of holding a given asset over various lookback periods utilizing: Skew and term structure of the underlying option market. Metrics of risk-adjusted performance like the agony/ecstasy ratio, Calmar ratio, max favourable move vs max adverse move, etc.
When does this break? One of the main assumptions is that volatility pricing is an indicator of sentiment and uncertainty. Our sentiment indicator would have done poorly in the late 90s. Prices were rising and sentiment was great, but volatility was also rising.
Similar to trend performance testing, we split indices into buy and sell group if the sentiment indicator was positive or negative.
The indicator works well from 1 week up to 3 months. Similar to trend, it captures a significant portion of its return by avoiding large drawdowns and capturing strong rebounds.
VP Analogs: the theory and the backtests
This is a quantitative expression of the old investor adage that “history doesn’t repeat, but it rhymes”. Today’s price action can often spur memories of historical episodes.
Our Analog indicator looks for patterns in historical price data and takes the subsequent return as a base rate. This is effectively a systematic and quantitative test of technical analysis.
Dynamic Time Warping gives some leeway in how the time axis is matched. We compare across the entire price history to find the most similar price patterns (top right chart shows the analogs for April 2020). The forward returns from the analogs form the base case (bottom chart).
When does this break? Using historical analogies implies the future will be like the past. This model will fail if there are significant and unprecedented changes in the macro regime
The tests presented here use 30 trading days for our price analogs and 2-week forward expectations.
It works better on commodities, but within each asset class there are some indices where this has not worked well historically.
Analogs assume the future can be represented by the past. One of our major concerns with this indicator is the performance on bond prices. There is very little data covering periods with low rates, or structurally rising inflation.
Putting it together
Combining Trend, Sentiment and Analog indicators adds robustness. The more signals that are positive, the better the forward returns (top right chart).
When all indicators are bearish, the tactical environment is very poor. This is consistent across the 1-week to 1-quarter horizon. The 1-month horizon does best across asset classes.
We create a sell regime if all 3 indicators are negative. Otherwise it is in the buy regime.
The main message is that the risk/reward is very low during periods when indicators are all negative.
Get the full picture at variantperception.com