Price forecasting & anomaly detection for agricultural commodities in India

Forecasting spot prices of agricultural commodities in India: Application of deep‐learning models

“…LSTM, a special type of RNN, has been extensively used for sequencing learning tasks, such as handwriting recognition and speech recognition (Graves et al, 2009;Graves, Mohamed, & Hinton, 2013;Hochreiter & Schmidhuber, 1997), and estimating the impact of incorporating news for specific companies (Siah & Myers, 2016). Madaan et al (2019) analyzed time-series mandi prices and arrival volumes of onion and potato by considering exogenous variables to the SARIMA and found it more accurate in predicting prices compared with the LSTM. Fisher and Krauss (2018) deployed LSTM for predicting out-of-sample directional movements for the stocks of the S&P 500 from 1992 until 2015.…”

Speculation in Delhi potato wholesale markets, 2007–2019: Causal connections of prices and arrival quantities

“…In addition to speculation in potato prices, onion prices are causing concern in India. Concerning these two vegetables of great political interest, the potato markets may be assumed to be competitive where hoarding and other speculative periods are received from the hands of millions independent agents, acting as a herd, whereas the onion markets are known for their high degree of cartelisation and price manipulation aimed to prevent entry of new players into the network (Fuglie, 1999;Madaan et al, 2019). Comparing results of this study to reactions by policymakers to speculation as covered by media, at least in regard to statistical significance, buffer stock policies appear promising in quieting demand-side speculation in potato prices (PTI, 2015).…”

“…This problem restricts studies in finance and other fields rich in high-frequency data, the information criteria are simply myopic to needed real-time adjustment lengths in potato markets in India. An early warning system needed 2-3 weeks of data for its over 60% accuracy in classifying between whether hoarding or weather affected prices, whereas forecasting accuracy still remained below 70% beyond few weeks when potato already requires cold storage (Madaan et al, 2019). It seems natural to use a real lag which approximately corresponds to Indian conditions and perishing of potato stocks without cold storage (Fuglie, 1999;Madaan et al, 2019) together with sensitivity analysis, acknowledging that adjusting reactions in physical markets are rarely immediate but require some time.…”

“…An early warning system needed 2-3 weeks of data for its over 60% accuracy in classifying between whether hoarding or weather affected prices, whereas forecasting accuracy still remained below 70% beyond few weeks when potato already requires cold storage (Madaan et al, 2019). It seems natural to use a real lag which approximately corresponds to Indian conditions and perishing of potato stocks without cold storage (Fuglie, 1999;Madaan et al, 2019) together with sensitivity analysis, acknowledging that adjusting reactions in physical markets are rarely immediate but require some time. The selected window size corresponds to maximal storage time of up to 5-6 months in India (Wustman et al, 2011;Madaan et al, 2019, the latter study also uses 6-month validation periods for anomaly detection).…”

“…It seems natural to use a real lag which approximately corresponds to Indian conditions and perishing of potato stocks without cold storage (Fuglie, 1999;Madaan et al, 2019) together with sensitivity analysis, acknowledging that adjusting reactions in physical markets are rarely immediate but require some time. The selected window size corresponds to maximal storage time of up to 5-6 months in India (Wustman et al, 2011;Madaan et al, 2019, the latter study also uses 6-month validation periods for anomaly detection). ADF testing rejects the unit root null hypothesis for both price and arrival quantity series even at 1% risk level using the chosen lag length.…”