Investment Research: AI-based forecasting system as an integrated DSS

Using I Know First AI-based forecasting system as an integrated DSS within a client’s investment or advisory process – case study with a bank

Key results

Applying screens derived from the I Know First AI-based forecasting system to recommendations list of a very well-known investment research company (=: “InvResCo’”) results in improved trade and portfolio performances:

On the trade level

  • the average return per trade increases from 1.9% to 2.6% – relative Delta of +36%
  • the winning rate increases from 60% to 63% – relative Delta of +5%

and the final recommendation list becomes shorter and clearer.

On the portfolio level

  • the total return (TR) increases from 29% to 35% – relative Delta of +20%
  • and the alpha against the SPY from 2% to 7%.

 

Case study – custom list of symbols based on InvResCo’s recommendation

In this document, an example on the utilization of a customized I Know First’s AI-based forecasting report is presented. The project was conducted on the request of a leading bank in Israel, primarily interested in improving its advisory services to investing clients.

The subject of the analysis are InvResCo’s highly rated companies covered by I Know First. An updated list of stocks selected by IRC is provided to the bank for its advisory activities each month of which a majority is covered by the I Know First forecasting system. From 123 unique stocks, each of which appears at least in one of the InvResCo files during the period March 2016 – March 2017 (13 months), I Know First covers 102 – these are U.S. companies with enough historical data. Those 102 stocks have been subject of the analysis, since this is the set on which I Know First adds value.

 

Analysis and back-testing approach

The benchmark has been set to 102 Stocks that are covered by I Know First and appear on a InvResCo list.

The InvResCo list is updated once a month, and is received by the bank usually in the second week of each month, but no later than the 15th of each month. Thus, the 1st, 7th and 15th of each Month have been selected as representative dates for recommendation update and/or portfolio rebalancing. In case a date falls on the weekend or a non-trading day in the U.S., the next trading day is taken instead. When applying I Know First’s algorithmic screen, the most recent forecast up to this date is used.

For the benchmark, all of the InvResCo stocks included in the current list and covered by I Know First’s algorithm are bought (equally weighted) and held until the next month.

In parallel, every month, from the list of 60-80 stocks provided by InvResCo, <=10 Stocks are selected to buy relying on I Know First’s predictions and the respective screens explained below.

The results “InvResCo” vs. “I Know First Screen of InvResCo” are then compared a) on a trade level and b) on a portfolio level. The first analysis applies to the financial advisory services the bank offers to individual clients, the latter in the context of launching and offering an investment vehicle (fund or other).

 

IKF Screen1: InvResCo list -> stocks covered by I Know First -> 35 most predictable stocks (14-days Predictability) -> 10 highest signals (14-days signal)

IKF Screen2: InvResCo list -> stocks covered by I Know First -> 35 most predictable stocks (1-month Predictability) -> 10 highest signals (1-month signal)

IKF Screen3: InvResCo list -> stocks covered by I Know First -> 35 most predictable stocks (3-month Predictability) -> 10 highest signals (3-month signal)

IKF Screen4*: InvResCo list -> stocks covered by I Know First -> 35 most predictable stocks (3-month Predictability averaged over the last 5 days) -> 10 highest signals (3-month signal averaged over the last 5 days). Idea is to have a more robust/persistent 3-month forecast through the averaging over the last five trading days.

IKF Screen5*: InvResCo list -> stocks covered by I Know First -> 35 most predictable stocks (both 1-month & 3-month Predictability averaged over the last 5 days) -> 10 highest signals (both 1-month and 3-month signal averaged over the last 5 days). Idea is to have a more robust/persistent 1-month and 3-month forecast through the averaging over the last five trading days.

Other screens are the respective combinations, that might result in less than 10 stocks per month, due to the intersection of the filtering criteria.

 

Performance results and comparison

Performance is analysed and compared to the benchmark on the trade and portfolio levels.

Trades analysis: March 2016 – April 2016

Below the average performance numbers of the strategies across the rebalancing dates are presented: on the trade level and on the monthly portfolio level.

*Note, that the analysis was performed based on the data available since 01.03.2016 and thus no rolling average of the recent forecasts until 01.03.2016 was possible.

One can see that applying I Know First’s signals as an additional algorithmic forecasting screen to InvResCo’s recommendation list improves the average trade return from 1.9% to 2.6% (rel. = +36%) and the winning rate of trades on average from 60% to 63% (rel. = +5%).

 

Portfolio analysis: March 2016 – April 2016

The added value of I Know First’s system can also be measured by considering a monthly rebalanced portfolio, constructed using the above-mentioned screens.

One can see in the table below that by applying the I Know First Screens higher returns and alpha are generated against the SPY than by buying the default InvResCo portfolio: 35% vs 29% (rel. = +20%) and 0.07 vs 0.02, respectively.

Note, that for the simulation the max weight per position was limited to 10% as this is a common requirement of mutual funds. Thus, the intersected screens which yield less than 10 trades per month on average are underinvested and deliver respectively lower returns than without the position size limit. All portfolio holdings are equally weighted.

 

 

Trade analysis: March 2016 – April 2016

 

Portfolio analysis: March 2016 – April 2016