Quantitative Price Prediction Model

Timing is of paramount importance when purchasing an item. If buying is off by a day, a week or a month, this can make the difference between a bargain or not. The Momentum Term Strength Index quantitative model (MTSI) ascertains whether a product is cheap or expensive to buy at the present price levels. It is utilized in conjunction with state of the art machine learning models.

MTSI is a new and exciting proprietary technical indicator. It is combined with convolutional and recurrent neural network technology to arrive at price prediction. To our knowledge, this on-line price bargain hunting device is not available by any other providers. It is our belief that this site will become an essential tool when purchasing items via the internet and will enable clients to find good value products and save money.

Product prices move in a series of peaks and troughs. The direction of these picks and troughs define the trend of the price. Upward trends have rising peaks and troughs. Downward trends have falling peaks and troughs. A trend has three directions, upward, when demand outweighs supply, downward, when supply outweighs demand, or sideways when demand and supply are in equilibrium
, John J. Murphy.

MTSI anticipates, to a degree, the trading behavior and captures the subsequent price action of two of the most popular quant strategies, namely, directional and contrarian:
  1. The directional is, perhaps, the most widespread strategy. These models go with the trend in prices. They find bargains if prices carry on moving in the same direction.
  2. Contrarian models go against the momentum or trend in prices. They profit when the price reverses in the opposite direction; usually at some important level.
The strength of the prevailing momentum or trend and its duration are two of the factors taken into account to produce buying signals.

MTSI rates the strength of trends, on a scale of 0 to 100. It accounts for timing and has the ability to handle trends in multiple time dimensions. It utilizes price and time-dependence, over varying time horizons for each product. It is capable of deciphering a variety of 'price frequencies'. Price oscillations, with differing time lengths, termed paths or dimensions, emulate the price action of the products. Paths are independent of one another.

The MTSI output is then combined with the output generated when product price history data is fed into:

  1. A Convolutional network. These networks are inspired by biological processes in that the connectivity pattern between elementary units resembles the organization of the animal visual cortex. The visual cortex of the brain processes visual information.

    This technology is applied to a recommender engine which is able to predict the product subsequent price action. Recommender systems produce a recommendation through content-based filtering. Content-based filtering utilizes a series of discrete characteristics of a product in order to recommend a future trend. This approach generates a probable future price for each product.
  2. A Recurrent neural network. In this network the connections between nodes form a directed graph, along a sequence. This graph allows it to exhibit temporal dynamic behavior.

    Initially, the algorithm is encoded with the neural network weights. Each weight encoded is assigned a respective weight link. The training set is presented to the network, which propagates the input signals forward. The mean-squared-error is returned and drives the selection process, thus resulting in a price prediction for a specific product.